FORMULATION OF ENGINEERED PARTICULATE SYSTEMS FOR CHEMICAL MECHANICAL POLISHING APPLICATIONS By GUL BAHAR BASIM A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2002
Copyright 2002 by Gul Bahar Basim
This study is dedicated to my parents Muzehher and Nahit Basim.
ACKNOWLEDGMENTS I would like to express my gratitude and appreciation to my advisor, Dr. Brij Moudgil, for his valuable guidance and support throughout this study. His sincere dedication to science, discipline in conducting research and attention to detail made significant contributions to this dissertation. I would also like to acknowledge the other members of my advisory committee, Dr. Rajiv Singh, Dr. Hassan El-Shall, Dr. Dinesh Shah and Dr. Wolfgang Sigmund, for their time, useful discussions and guidance. I gratefully acknowledge the National Science Foundationâ€™s Engineering Research Center for Particle Science and Technology for financially supporting this research. Special thanks go to ERC staff members Mike Beasley, Gill Brubaker, Genena Blanchette, Rhonda Blair, Cheryl Bradley, John Henderson, Lenny Kennedy, Sophie Leone, Byron Salter, Gary Scheiffele, Nancy Sorkin, ERC faculty members, Kevin Powers, Yakov Rabinovich, Abbas Zaman, Brian Scarlet and ERC administrators Anne Donnelly and Craig Pledger, who made this work possible. I would also like to extend my gratitude to my mentor, Dr. Reg Davies, who has set a great example of lifetime success in research and engineering and to Dr. Mohsen Khalili who guided my research at DuPont. My colleagues, Joshua Adler, Scott Brown, Won-Seop Choi, Madhavan Esayanur, James Kanicky, Seungh-Mahn Lee, Uday Mahajan, Byron Palla, Pankaj Singh, Kuide Qin, Ivan Vakarelski and Suresh Yeruva, who have contributed to this research by their valuable discussions and friendship, are also gratefully acknowledged. My iv
undergraduate students Hema Patel, Udeni Fernando, Jason Spuhler, Ketan Chadesama and Ashish Daruka deserve special thanks for their valuable assistance throughout this work. Finally, I would like to thank to my husband, Kutsal Dogan, for his invaluable support and understanding throughout my Ph.D. This work would not have been possible without him. I would also thank to my parents, Muzehher and Nahit Basim, for their continuous support in developing my career and guidance in life. v
TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES.............................................................................................................ix LIST OF FIGURES.............................................................................................................x ABSTRACT.....................................................................................................................xiv CHAPTERS 1. INTRODUCTION.........................................................................................................1 2. OVERVIEW OF CHEMICAL MECHANICAL POLISHING....................................8 Motivations for Chemical Mechanical Polishing Process...........................................8 Multilevel Metallization (MLM).........................................................................10 Planarization........................................................................................................12 The CMP Process.......................................................................................................13 Surface to be Polished.........................................................................................14 Polishing Pad.......................................................................................................17 Polishing Slurry...................................................................................................21 Effect of slurry chemistry on CMP performance..........................................22 Metal CMP.................................................................................................22 Silica CMP.................................................................................................28 Effect of slurry particulate properties...........................................................33 Modeling Efforts in CMP..........................................................................................37 Scope of the Dissertation...........................................................................................40 3. DETECTION AND IMPACT OF HARD AGGLOMERATES IN CMP...................42 Introduction................................................................................................................42 Experimental..............................................................................................................44 Results and Discussion..............................................................................................45 Detection of Coarser Particles in CMP Slurries..................................................45 Surface Roughness and Critical Defect Analysis on the Wafer Surfaces in the Presence of Coarser Particles During CMP..................................................48 Material Removal Rate Response in the Presence of Coarser Particles.............51 Summary....................................................................................................................55 vi
4. FORMATION AND IMPACT OF SOFT AGGLOMERATES IN CMP...................57 Introduction................................................................................................................57 Experimental..............................................................................................................58 Results and Discussion..............................................................................................60 Effect of Dry Aggregates....................................................................................60 Effect of Polymer Mediated Soft Agglomerates (Flocs).....................................62 Effect of Salt Coagulated Agglomerates.............................................................66 Summary....................................................................................................................70 5. STABILIZATION OF CMP SLURRIES BY CONTROLLING INTERACTION FORCES..................................................................................................................72 Introduction................................................................................................................72 Experimental..............................................................................................................73 Results and Discussion..............................................................................................77 Dispersion of CMP Slurries by Controlling Particle-Particle Interactions.........77 Control of Pad-Particle-Substrate Interactions for Surfactant Mediated Slurries80 Effect of self assembled surfactant aggregates on particle-substrate interactions...............................................................................................81 Force per particle during polishing...............................................................83 Surface lubrication effect on particle-substrate interactions.........................87 In-situ friction force measurement................................................................87 Friction force measurements with AFM.......................................................89 Modification of the Interaction Forces by Calcium Addition.............................92 Summary....................................................................................................................96 6. DESIGN CRITERIA FOR OPTIMALLY PERFORMING CMP SLURRIES..........98 Introduction................................................................................................................98 Experimental..............................................................................................................99 Results and Discussions...........................................................................................101 Material Removal Rate Response.....................................................................101 Effect of slurry particle size and concentration on material removal rate..102 Effect of applied head pressure on material removal..................................113 Surface Quality Response.................................................................................114 Optimization......................................................................................................117 Summary..................................................................................................................121 7. SUMMARY AND SUGGESTIONS FOR FUTURE WORK..................................124 Summary..................................................................................................................124 Suggestions for Future Work...................................................................................128 APPENDIX CENTRAL COMPOSITE DESIGN AND RESPONSES...............................................131 vii
LIST OF REFERENCES.................................................................................................133 BIOGRAPHICAL SKETCH...........................................................................................141 viii
LIST OF TABLES Table page 1.1. High-performance logic technology requirements......................................................4 1.2. National technology roadmap for semiconductors......................................................4 3.1. Detection limits of Coulter LS 230 for Rodel 1200 slurry spiked with 0.5, 1.0 and 1.5 m Geltech particles.............................................................................................47 3.2. Summary of the polishing results for the Rodel 1200 slurry spiked with 0.5, 1.0 and 1.5 m Geltech silica particles at the established detection limits.............................52 4.1. Slurry performance summary in the presence of soft agglomerates...........................60 5.1. Summary of polishing performance of the baseline (with and without NaCl), C 12 TAB and C 8 TAB mediated slurries in the presence of 0.6 M NaCl.....................78 5.2. Summary of polishing performance of the baseline (with and Without CaCl 2 ), and C 12 TAB mediated slurries in the presence of 0.24 M CaCl 2 ......................................93 6.1. Selected design levels for the three-factor rotatable central composite design.......100 ix
LIST OF FIGURES Figure page 1.1. Schematic representation of chemical mechanical polishing (CMP) process. (a) Side view; (b) Top view............................................................................................2 2.1. Multilevel metallization. a) 3-D, low-angle scanning electron micrograph of a portion of partially completed SRAM array containing six-device memory cells; (b) Cross section with silica dielectric and aluminum metallization.......................11 2.2. Degrees of surface planarity....................................................................................12 2.3. Schematic representation of copper dishing and SiO 2 erosion ...............................16 2.4. IC 1000/Suba IV stacked polishing pad SEM cross-section...................................18 2.5. IC-1000 polishing pad surface (a) SEM micrograph (200x); (b) AFM micrograph and cross-section.....................................................................................................19 2.6. SEM micrograph of Suba IV pad surface (150x). (a) Before polishing; (b) After polishing..................................................................................................................20 2.7. SEM micrograph of fixed abrasive pad surface.......................................................20 2.8. Tungsten polishing in the absence and presence of oxidizer in the CMP slurry 23 2.9. Mechanism of tungsten polishing and planarization in the presence of passive oxide layer...............................................................................................................24 2.10. Current recorded during wear experiments of tungsten for anodic applied potential (+2V) in 0.5 M H 2 SO 4 for a sphere motion cycle of 0.2 seconds ..........................25 2.11. Thin oxide film behavior as a function of the P-B ratio (a) Porous oxide; (b) Protective oxide; (c) Film fracture..........................................................................27 2.12. Vickers indentation hardness of silica glass as a function of water content ..........31 2.13. Scanning electron micrographs of Vickers indentation of silica glass with different water contents in toluene (a) ~0 wt% water; (b) ~6.11 wt% water; (c) ~8.65 wt% water........................................................................................................................32 x
2.14. Material removal rate of silica as a function of the solids loading of silica slurries for different particle sizes........................................................................................34 2.15. Schematic representation of material removal mechanisms. (a) Contact area mechanism; (b) Indentation volume based mechanisms.........................................35 3.1. Particle size distributions for Rodel 1200 slurry spiked with Geltech 1.5m particles...................................................................................................................46 3.2. AFM pictures and cross sections for the wafers polished with Rodel 1200 slurry before and after spiking with coarser size particles. (a) Rodel 1200 baseline slurry; (b) Rodel 1200 slurry spiked with 1.1 wt% Geltech 0.5 m particles; (c) Rodel 1200 slurry spiked with 1.1 wt% Geltech 1.5 m particles....................................49 3.3. Cross-section analysis for the wafer polished with spiked Rodel 1200 slurry. (a) Slurry spiked with 0.3 wt % Geltech 1.0 m particles; (b) Slurry spiked with 1.3 wt% Geltech 1.0 m particles.................................................................................50 3.4. Polishing rates for the baseline and spiked Rodel 1200 slurries..............................53 3.5. Change in polishing mechanism for the Rodel slurries spiked with Geltech coarser particles...................................................................................................................54 4.1. Particle size distribution of the baseline and soft agglomerated slurries. The soft agglomerates were prepared by dry aggregation, polymer flocculation and salt coagulation methods at 5-10 m size range............................................................59 4.2. AFM images of the silica wafers after CMP. (a) Baseline 0.2 m 12 wt% monosize silica slurry; (b) Slurry with dry aggregates...........................................61 4.3. AFM images of the silica wafers after CMP. (a) Slurry flocculated using 0.5 mg/g PEO; (b) Slurry dispersed in the presence of 0.5 mg/g PEO..................................62 4.4. Particle size distribution of the baseline and polymer flocculated slurries. The size distribution of the flocculated slurries shifted down after polishing. Slurries stirred long enough in the presence of PEO dispersed to the original size distribution of the baseline slurry....................................................................................................63 4.5. AFM friction coefficient measurements for the baseline and PEO containing solutions ..................................................................................................................65 4.6. Particle size distribution of the baseline and salt coagulated slurries. The size distribution of the slurry with 0.2 M NaCl did not change significantly. Addition of 0.6 M NaCl coagulated the slurry. Size distribution of the 0.6 M NaCl slurry before and after polishing remained the same.........................................................67 4.7. AFM images of the silica wafers after CMP. (a) Polished with 0.2 M NaCl containing slurry; (b) polished with 0.6 M NaCl containing slurry......................69 xi
5.1. Schematic illustration of AFM force measurement technique................................74 5.2. Schematic illustration of in-situ force measurement technique...............................75 5.3. Schematic representation of the FTIR-ATR technique used to measure the percent pad contact as a function of down load...................................................................76 5.4. Schematic representation of the particle-particle and particle-substrate interactions for the silica-silica polishing systems in the presence of self assembled surfactant aggregates................................................................................................................80 5.5. Maximum repulsive force response of C12TAB, C10TAB and C8TAB surfactants at 32, 68 and 140 mM concentrations in the presence of 0.6 M NaCl at pH 10.5 obtained with AFM using 1.5 m particle attached to tip.......................................81 5.6. FTIR-ATR spectra for IC-1000 polishing pad. (a) Porous pad contact as a function of applied load; (b) Pore free, 100 % pad contact as a function of applied load..........................................................................................................................84 5.7. Percent IC 1000 pad contact as a function of applied load......................................85 5.8. SEM image of the IC-1000 polishing pad surface after polishing with 0.2m size 12 wt% polishing slurry..........................................................................................86 5.9. In-situ friction force and material removal rate responses of the baseline slurries and the slurries containing C 12 TAB, C 10 TAB and C 8 TAB surfactants at 32, 68 and 140 mM concentrations in the presence of 0.6 M NaCl at pH 10.5........................88 5.10. AFM friction force measurements on silica wafer with 7m size particle attached to the tip. (a) Solutions containing C12TAB, C10TAB and C8TAB surfactants at 32, 68 and 140 mM concentrations without NaCl at pH 10.5; (b) In the presence of 0.6 M salt in the solution.........................................................................................90 5.11. FTIR-ATR adsorption spectra of the solutions in the presence of 32 mM C 12 TAB without salt and in the presence of 0.6 M NaCl and 0.24 M CaCl 2 ........................94 5.12. AFM force measurements for C 12 TAB mediated slurry in the presence of NaCl and CaCl 2 . (a) Repulsive force measurements; (b) Friction force measurements.......95 6.1. Central composite design for three factors............................................................100 6.2. Prediction profiles for material removal rate response..........................................101 6.3. Material removal rate surface response for 0.5 m particle size slurry.................102 6.4. Material removal rate analysis for 0.3, 0.5 and 0.8 m size slurries as a function of solids loading.........................................................................................................103 xii
6.5. Suggested particle-substrate interactions for 0.3 m size slurries as a function of solids loading with the AFM surface micrographs...............................................105 6.6. SEM images conducted on the pad surface after polishing with 0.5 %wt slurries. (a) 0.3 m; (b) 0.5 m; and (c) 0.8 m particle size.............................................106 6.7. Pad-particle-substrate interactions as a function of increasing slurry solids concentration.........................................................................................................107 6.8. SEM images conducted on the pad surface after polishing with 30 %wt slurries. (a) 0.3 m; (b) 0.5 m; and (c) 0.8 m particle size.............................................108 6.9. In situ friction force measurements of 0.3 m size slurries with material removal rate response as a function of solids concentration...............................................109 6.10. Schematics of the pad-particle-substrate interactions with the forces induces during polishing. Particle starts rolling when exceeds 36 o ...........................................110 6.11. Effect of head pressure on material removal rate for abrasive free and 0.3, 0.5 and 0.8 m size slurries................................................................................................114 6.12. Prediction profiles for surface roughness response...............................................115 6.13. Prediction profiles for maximum depth of the surface defects (pits or scratches) response.................................................................................................................115 6.14. AFM pictures and surface profiles for the wafers polished with 15 wt% slurries under 74N applied downforce. (a) 0.2 m size slurry; (b) 1.0 m size slurry....116 6.15. Optimal operational regimes as a function of solids concentration and particle size under different applied load conditions. (a) 54 N applied load; (b) 94 N applied load; (c) 76.2 N applied load.................................................................................120 6.16. CMP slurry design criteria.....................................................................................122 xiii
Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy FORMULATION OF ENGINEERED PARTICULATE SYSTEMS FOR CHEMICAL MECHANICAL POLISHING APPLICATIONS By Gul Bahar Basim August 2002 Chair: Dr. Brij M. Moudgil Department: Materials Science and Engineering Chemical mechanical polishing (CMP) is widely used in the microelectronics industry to achieve planarization and patterning of metal and dielectric layers for microelectronic device manufacturing. Rapid advances in the microelectronics industry demand a decrease in the sizes of the devices, resulting in the requirement of a very thin layer of material removal with atomically flat and clean surface finish by CMP. Furthermore, new materials, such as copper and polymeric dielectrics, are introduced to build faster microprocessors, which are more vulnerable to defect formation and also demand more complicated chemistries. These trends necessitate improved control of the CMP that can be achieved by studying the slurry chemical and particulate properties to gain better fundamental understanding on the process. In this study, the impacts of slurry particle size distribution and stability on pad-particle-surface interactions during polishing are investigated. One of the main problems in CMP is the scratch or pit formation as a result of the presence of larger size particles in xiv
the slurries. Therefore, in this investigation, impacts of hard and soft (transient) agglomerates on polishing performance are quantified in terms of the material removal rate and the quality of the surface finish. It is shown that the presence of both types of agglomerates must be avoided in CMP slurries and robust stabilization schemes are needed to prevent the transient agglomerate formation. To stabilize the CMP slurries at extreme pH and ionic strength environments, under applied shear and normal forces, repulsive force barriers provided by the self-assembled surfactant structures at the solid/liquid interface are utilized. A major finding of this work is that slurry stabilization has to be achieved by controlling not only the particle-particle interactions, but also the pad-particle-substrate interactions. Perfect lubrication of surfaces by surfactants prevented polishing. Thus, effective slurry formulations are developed by studying the frictional forces, which are representative of the particle-substrate interactions, while achieving stability by introducing adequate interparticle repulsion. Finally, optimal slurry particulate properties are examined by analyzing the material removal mechanisms for silica-silica polishing. Based on the reported findings, a slurry design criterion is developed to achieve optimal polishing performance. xv
CHAPTER 1 INTRODUCTION Chemical mechanical polishing (CMP) combines chemical and mechanical interactions to planarize surfaces using a polishing slurry that is made of submicron size particles and chemicals. The other main components of the CMP process are the surface to be polished and the polymeric polishing pad. Slurry chemical components interact with the material to be polished and alter its mechanical properties by creating a chemically modified surface layer. Simultaneously, slurry abrasives mechanically interact with the chemically modified substrate surface, resulting in material removal. The polymeric polishing pad is responsible for carrying the slurry underneath the substrate and transferring the applied head pressure to the abrasive particles. Figure 1.1 schematically illustrates the CMP process. The material to be polished (wafer) is mounted on a rotating polishing head, which comes in contact with a rotating platen carrying the polymeric pad. Polishing slurry flows between the pad and the substrate surfaces resulting in planarization. The CMP process has been used to polish a variety of materials to produce optically flat and mirror finished surfaces [Ste97]. It has been adopted for preparation of optically flat and damage-free surfaces for glasses and semiconductors. In the microelectronics industry, CMP is used for the planarization of interlayer dielectrics (ILD) and metals, to form interconnections between devices during multilevel metallization (MLM). The main advantage of the CMP process is that it enables global 1
2 (a) (b) Figure 1.1. Schematic representation of chemical mechanical polishing (CMP) process. (a) Side view; (b) Top view [Oum99]. planarization, which is essential in building sub-0.5 m MLM. Planarization is achieved in CMP by maintaining high removal rates at the high surface features, while low surface features are removed very slowly [Car90]. This difference in polishing rate is achieved by the use of a rigid polishing pad that exerts relatively larger force onto the high features but does not conform to the wafer surface. Lower forces exerted on the low features results in their minimal removal leading to planarization. By accomplishing global planarization, CMP significantly reduces nonplanarity defects such as poor step coverage and metal stringers, which form when the thick metal film at the edge of a step is not completely etched. The CMP process is also cost effective and provides alternative means of patterning metal, eliminating the need for the reactive ion etching for difficult-to-etch metals. Consequently, it increases the die yields and decreases the die cost [Ste97]. However, improper CMP may result in other types of defects such as scratching or pitting, stress cracking, delamination of weak interfaces, corrosive attack from the
3 slurry chemicals and residual abrasive particles. These defects may also lead to the production of imperfect microprocessors. Formation of defects becomes more critical as the device dimensions decrease. Fabrication of 90 nm devices is expected by 2004 with 37 nm wide gates, which will be made of only three to four atomic layers of dielectrics [Pet02a]. Therefore, a very thin layer of material removal with minimal number of defects has to be achieved by CMP. Furthermore, 300 mm wafers are currently in production yielding a higher number of chips as compared to previously used 200 mm wafers, resulting in higher loses than before, in case a wafer is discarded [Bra02]. Emerging introduction of copper and low-k dielectrics to build faster and more reliable devices raises more challenges in the efforts of reducing defects in CMP, since these are softer materials and hence more vulnerable to defect formation [Bra01]. Tables 1.1 and 1.2 summarize the future trends in manufacturing of high performance microelectronic devices. It is seen that the device dimensions keep shrinking while the wafer sizes continue to increase, indicating that significant improvements will be needed in CMP performance to keep up with the industrial demands. This requires a fundamental understanding of the chemical and mechanical aspects of the pad-slurry-substrate interactions during polishing. Despite the need for improved understanding of the CMP process at atomic level, the emphasis remains to be the adjustment of the operational variables such as the applied head pressure and the platen velocity. Yet, the material removal takes place primarily as a result of the chemical and mechanical interactions provided by the polishing slurry. It is therefore necessary to investigate the role of slurry chemical and particulate properties in designing effective CMP processes. Slurry chemistry modifies the properties of the
4 Table 1.1. High-performance logic technology requirements. 2001 2002 Year 2003 2004 2005 2006 2007 Technology node (nm) 130 90 65 Physical gate length (nm) 65 53 45 37 32 28 25 Equivalent oxide thickness (EOT) (nm) 1.3-1.6 1.2-1.5 1.1-1.6 0.9-1.4 0.8-1.3 0.7-1.2 0.6-1.1 Table 1.2. National technology roadmap for semiconductors. 1995 1998 Year 2001 2004 2007 2010 Minimum feature size (m) 0.35 0.25 0.18 0.13 0.1 0.07 Maximum substrate diameter (mm) 200 200 300 300 400 400 Number of metal levels DRAM Microprocessor 2 4-5 2-3 5 3 5-6 3 6 3 6-7 3 7-8 Interconnection metal Al, Cu Al, Cu Al, Cu Cu, Al Cu, Al Cu, Al ILD Dielectric Constant 3.9 <3 2.5 2 1-2 1-2 Source: [ Pet02 ] Source: [ Ste97 ] surface to be polished [Kau91]. The mechanical interactions, on the other hand, vary depending on the slurry particle size and concentration, since these factors alter the load applied per particle. By further understanding of the fundamentals of material removal mechanisms at single particle-substrate interaction level, methodologies can be developed to formulate efficient polishing slurries.
5 Slurry particle size distribution also plays a very important role in CMP, particularly in terms of defect formation. It has long been suggested that the presence of larger size particles in the slurries is the major reason for the scratches [Sch95]. Although there are several methods applied in practice to monitor the slurry particle size distribution and detection of larger size particles, their impact on CMP performance has not been quantified [Poh96, Nag96, Duk98]. Furthermore, factors responsible for robust slurry stability need to be investigated since the formation of transient agglomerates during polishing may as well be the reason for defect formations. In summary, engineering the slurry particulate properties appears to be the crucial factor in developing a consistently high performing polishing system. A synopsis of the efforts constituting this study is organized as follows. Chapter 2 reviews the literature on the CMP process. Initially, motivations in the microelectronics industry for conducting CMP are discussed. Furthermore, roles of the three main components in polishing and planarization: surface to be polished, polishing pad and the polishing slurry are introduced. The previous findings on the impact of slurry particulate properties, such as the particle size and particle size distribution, are summarized. In addition, effects of slurry chemistry on modifying the substrate surface are reviewed for the metal and silica-silica polishing systems as a major requirement for providing defect free surface finish with acceptable material removal rate. The Silica polishing is studied in this investigation as a model system being the most commonly used dielectric material in microelectronics manufacturing and requiring the least complex slurry chemistry. Mechanisms of material removal and defect formation are also discussed in this chapter. Finally, the modeling efforts in predicting polishing
6 performance are summarized and discussed in terms of their effectiveness in evaluating not only the material removal rate response but also the surface quality of the polished substrate. New approaches for developing more effective predictive methodologies are introduced. Chapter 3 focuses on the role of hard, larger size particles in CMP slurries. Their impact on the polishing performance is quantified in terms of the material removal rate response and surface quality of the wafers. Chapter 4 investigates the impact of transient soft agglomerates on polishing performance. Soft agglomerates are introduced into the polishing slurries by slurry chemical modifications and dry aggregation methods. Behavior of transient agglomerates during CMP is evaluated by monitoring the change in polishing performance as a function of variation in slurry particle size distribution before and after polishing. Chapter 5 discusses stabilization of the CMP slurries utilizing the repulsive force barriers created by the self-assembled surfactant structures. Slurry stability is studied in terms of not only the particle-particle interactions but also the pad-particle-substrate interactions. Mechanisms of material removal are investigated for the surfactant-mediated slurries by studying their impact on the frictional forces encountered during polishing. The criteria for effective slurry formulation to achieve stability, while providing sufficient material removal rates with minimal defects, are defined. Chapter 6 concentrates on the design criteria for optimally performing CMP slurries in terms of the slurry particulate properties. Particularly, slurry particle size and solids concentration effects are investigated as a function of the applied head load. Furthermore, slurry particulate properties are optimized as a function of the applied down
7 load in this chapter. Chapter 7 summarizes the findings of this study with the suggestions for future work.
CHAPTER 2 OVERVIEW OF CHEMICAL MECHANICAL POLISHING In this chapter a brief description of the CMP process is presented starting with the reasons fort the planarization requirement in microelectronic device manufacturing. The there main components of the CMP process, surface to be polished, polishing pad and slurries are introduced and their major roles in polishing are discussed. Furthermore, current practice for designing highly efficient polishing slurries are analyzed to identify specific research needs for developing reliable predictive methodologies for the CMP applications. Motivations for Chemical Mechanical Polishing Process The continuous demand in the semiconductor industry for developing faster and more functional processors has resulted in fabrication of increasingly complex, dense and miniaturized devices and circuits [Ste97]. As the device dimensions are scaled down, the integrated circuits (IC) face the interconnect delay problem associated with the capacitance (C) and resistance (R) of the metal lines. Most commonly, the interconnect delay is measured as RC time delay, which is defined as the time it takes for the voltage at one end of a metal line to reach 63% of its final value when a step input is presented at the other end of the line [Ste97]. To define the relationship of the RC time delay to the circuit design variables, resistance and capacitance of the metal lines should be defined initially. The resistance of a metal line is expressed as given in Equation 2.1, AlwdlR (2.1) 8
9 where is the metal resistivity and l, w, d, and A are the length, width, thickness, and cross sectional area of the line respectively. Capacitance of the metal line, is given by, twl C (2.2) where is permittivity (equal to the insulator dielectric constant, r , times the permittivity of free space o ) and t is the thickness of the insulator. Since the RC time delay is the direct product of capacitance and resistance of the metal line, it can be expressed as in Equation 2.3 by combining Equations 2.1 and 2.2, tdlRC2 (2.3) From this relationship, it is clear that to reduce RC time delay, , and l must be decreased, while d and t must be increased. Although the RC delay is not dependent on the line width directly, decreasing the feature size generally requires that the line thickness (d) and the inter level dielectric (ILD) thickness (t) be scaled down. Therefore, eventually the RC delay increases according to the relationship represented in Equation 2.3. In addition, the capacitance between the adjacent metal lines increases with the decreasing feature size. Especially below 0.5 m, a dramatic rise is recorded in RC time delay, which exceeds the advantage of decreasing device dimensions to decrease the device delay [Ste97]. As can be deduced from Equation 2.3, decrease in line length has the major impact on decreasing the RC delay since the delay increases with the square of the length. Thus, it is advantageous to avoid long metal lines in circuit designs. In addition, the width of the interconnections should be large enough to reduce line-to-line capacitance. These two requirements can be met by the multilevel metallization method, which is explained in the following section.
10 Multilevel Metallization (MLM) Figures 2.1 a and b illustrate the three-dimensional and cross sectional scheme of multilevel metallization [Sti97]. In MLM, metal interconnections span several planes isolated by the insulating dielectric layers, which are wired in the third dimension by vias in the dielectric planes [Mur93]. In addition to reducing the line length and increasing the interconnection width, the MLM design also enables routing the metal lines to the upper levels when the long lines cannot be avoided. To achieve MLM, several steps are required, including deposition, lithography, etching and planarization. Initially, the dielectric surface is planarized and high aspect ratio via holes are etched into it. Then these holes are filled with metal and the excessive metal is removed from regions other than the vias by polishing. On top of the vias, metal lines are deposited and patterned. Finally, the next layer of dielectric is deposited and planarized to build the subsequent level. The decreasing device dimensions and the increasing number of metal layers necessitate the improved planarization of the surfaces for MLM, since non-planarity of the surface becomes cumulative as the layers are built one on top of another. High topographic variation results in poor step coverage and electromigration problems [Ols93]. Furthermore, below the 0.5 m regime, the lithography tools require high numerical aperture lenses to print fine line dimensions. The depth of field of these lenses is approximately 270 nm across a 27 by 27 mm stepper field. Due to this finite depth, the optical steppers require less than 150 nm variation in topography across the stepper field. The uneven surface topography also causes uneven photoresist thickness resulting in overexposure at the thinner resist layers. Hence, effective planarization of the wafer surface must be achieved for MLM.
11 (a) (b) Figure 2.1. Multilevel metallization. a) 3-D, low-angle scanning electron micrograph of a portion of partially completed SRAM array containing six-device memory cells (insulating oxide removed); (b) Cross section with silica dielectric and aluminum metallization [Sti97].
12 Figure 2.2. Degrees of surface planarity [Ols97]. Planarization Figure 2.2 illustrates the three degrees of planarity of (i) no planarization, (ii) local planarization and, (iii) global planarization. Local planarity achieves locally flat surfaces but the overall surface height may vary across the die. In global planarization, on the other hand, the surface is flat across the entire wafer [Ols93]. There are a couple of techniques available to achieve planarization. The simplest approach is to flow the SiO 2 (dielectric) layer with a high temperature anneal. By doping SiO 2 with boron, the glass transition temperature is lowered, allowing flow at temperatures as low as 900 o C [Wol86]. However, flow anneals may only be performed before the first aluminum metal layer is deposited, since aluminum melts at the high anneal temperatures. Use of tungsten as the first level metal enables flow anneals since tungsten is stable at this temperature. The second alternative for planarization is dry etching or reactive ion etching (RIE). After the deposition of the oxide layer, a thick layer of photoresist is deposited on the surface, covering the steps and giving a flat finish. The photoresist etch
13 rate is matched to the oxide etch rate so that the surface can be etched by RIE process uniformly, and planarized by etching until the plane oxide layer is reached. This technique is disadvantageous, since often several repeats are required and the plasma used is composed of hazardous gases [Mur93, Ste97]. The third alternative is the chemical mechanical polishing (CMP), which is described in the following section. The CMP Process The main advantage of chemical mechanical polishing comes from its ability to achieve global planarization. Being able to achieve global planarity, it reduces the planarity related defects such as metal stringers and poor step coverage. The excessive film parts remaining from the earlier steps can also be removed during CMP. It creates an alternative to dry etching for materials like copper, which are not amenable to dry etch [Ste97]. In spite of its advantages, CMP has its own challenges to planarization including, (i) an entire new set of metrology tools for its integration into the IC manufacturing and, (ii) inadequate understanding of process fundamentals. It also results in new types of defects on the wafers such as scratches, pitting, stress cracking of the films, corrosive attack of the slurry chemicals, delamination of film interfaces and residual particles on the film surfaces [Lan92]. To increase CMP efficiency, it is necessary to acquire better understanding of the basis that will lead to enhanced process control and reduced number of defects. The CMP process consists of moving the surfaces to be polished against a polymeric pad that is used to provide support against the sample surface and to carry the polishing slurry. Polishing slurry is composed of submicron size particles and chemicals providing the chemical and mechanical interactions that enable material removal. In CMP, planarization is achieved by maintaining high removal rates at the high surface
14 features, while low surface features are removed relatively slowly [Car90]. To understand polishing and planarization, roles of the three main components of CMP, namely, the surface to be polished, polishing pad and the slurry, are outlined below. Surface to be Polished The successful application of the CMP process in silicon ICâ€™s was started with multilevel interconnection structures employing SiO 2 as the inter level dielectric, the chemical vapor deposited (CVD) tungsten as the via fill metal, and the sputtered aluminum as the planar interconnection metal [Moy89]. Initial process developments in CMP focused on the SiO 2 and tungsten layers [Utt91, Kau91]. Since then, the use of CMP has expanded to a large variety of metals including Al, Cu, Ta, Ti, TiN, W and their alloys and insulators such as SiO 2 and SiO 2 doped glasses, Si 3 N 4 , polymers and polysilicon. The main motivation for the search for new materials in the microelectronics industry has come from the requirement of reducing the RC time delay. As discussed earlier, RC time delay can be reduced by decreasing the resistivity of metal and the dielectric constant of the ILD. The most commonly used metals in IC manufacturing are the aluminum alloys. While aluminum is a good conductor, its resistivity is 2.66 -cm. There are other metals available with lower resistivities such as copper, silver and gold. Gold is favorable for its high corrosion and electromigration resistance; however, it shows only a marginal improvement in resistivity with a value of 2.35 -cm. The other alternative is silver with 1.59 -cm resistivity but it tends to diffuse into SiO 2 [McB86]. These two metals also dramatically affect the electronic properties of silicon due to the two deep energy levels introduced into the silicon bandgap [Sze81]. Finding barriers to
15 silver diffusion into SiO 2 and silicon has proven difficult [Mur91]. In addition, silver has the disadvantage of poor electromigration resistance because of its low melting point [Li94]. Therefore, copper appeared to be the most attractive candidate among all the other metals. Its resistivity is 1.67 -cm, which is 50% lower than the aluminum alloys. It also has a greater electromigration resistance than aluminum [Mur93], which allows for the higher current densities before the onset of electromigration [Li94]. Although copper also exhibits deep levels in silicon and its impurities diffuse in silica, there are several effective barrier materials available to prevent these effects, typically tantalum, tantalum nitride and silicon nitride [Wan94]. The other challenges for copper are its susceptibility to corrosion and unsuitability to dry etching. However, there are promising methods for passivation of copper as well as an alternative patterning methodology called dual damascene, in which trenches and holes are cut into dielectric and filled with copper [Ste97]. In damascene technology, first the vias and trenches are created and a TaN diffusion barrier is physical vapor deposited (PVD) in them followed by the PVD deposition of a copper seed layer. Afterwards, the bulk of the copper is deposited by electroplating and the surface is planarized by CMP. The copper process is well adopted at 0.18 m technology node and it remains the metal of choice for the future ICâ€™s [Sin02] In terms of the selection of low dielectric constant ILD, polymeric materials are promising candidates [Ste97]. The currently used ILDâ€™s are mostly chemical vapor deposited SiO 2 with r between 3.9 and 6.0 depending upon the doping (fluorine or boron) and water content of silica. While low r polymer materials show promise as ILDâ€™s, several issues must be resolved before their incorporation into the IC
16 manufacturing process [Tho94]. First, the selected polymers must be stable at the applied high processing temperatures. Furthermore, they have to be mechanically stable resisting to stress induced deformation and delamination. Their anisotropy in r and patterning are other concerns in their integration. SiLK by Dow Chemical and GX-3 by Honywell are some of the commercial polymeric dielectric products. The bulk of the low-k implementation is expected at 90 nm device node, where organosilicate and spin-on polymers will dominate at k=2.7 [Pet02b]. As new materials are introduced to IC manufacturing, their integration brings in new challenges. Currently, the copper-low k integration is being studied extensively to enable high yield and minimal defects during planarization [Pet02a]. One of the major challenges is the dishing and erosion problem for the copper interconnects (Figure 2.3) [Ste97]. This problem is especially severe where there is either a large difference in the properties of the two materials being polished (such as hardness), or when extremely tight dimensional control is required. Dishing occurs when wide and isolated features are exposed to CMP. If the polishing slurry has a chemical selectivity to the metal in trench Figure 2.3. Schematic representation of copper dishing and SiO 2 erosion [Ste97].
17 (Cu) or this material is softer than the selected dielectric (like copper and silica system), it may be polished much faster relative to the surrounding material. Dishing can lead to poor electrical contact in MLM structures. Erosion, on the other hand, takes place when the dielectric surrounding the metal features is polished along with the metal in the high pattern density areas. As it leads to thinning of the dielectric layer along with the metal, the planarity as well as the electrical properties of the MLM are degraded. The CMP process therefore, needs to be designed depending on the materials to be polished on the wafer surface. Polishing Pad The main function of the polishing pad is to transport the polishing slurry to the wafer surface and provide planarity as it interacts with the high level structures on the wafer. The surface structure and material properties of the polishing pads are important in determining the material removal and planarization ability of the CMP process. In general, polishing pads are made of polyurethane since chemistry of polyurethane allows the pad characteristics (such as hardness and porosity) to be tailored to meet specific material property needs in CMP [Jai94]. The harder and more incompressible the pad, the less it will bend and conform to the wafer surface resulting in less material removal at the recessed areas thus, increasing the planarization. However, some flexibility of the pad is also required to minimize any possible breakage of wafers during polishing. These demands led to the use of stacked pads. Figure 2.4 shows the schematic cross-section of Rodelâ€™s composite polish pad consisting of two pre-stacked films. The bottom pad film (Suba IV) is softer than the upper film. It is used to ensure that the composite structure conforms to the wafer shape. The top stiffer film (IC-1000) maintains a long planarization length when polishing patterned wafers, enabling global planarization.
18 Figure 2.4. IC 1000/Suba IV stacked polishing pad SEM cross-section (40 x) [Obe98]. The structure of the pad surface is important for the slurry transport. Figures 2.5 a and b show the SEM and AFM images of the IC-1000 polishing pad with the AFM cross section. The macro and micropores on the pad surface can be clearly seen in Figure 2.5 a, which help slurry transport. The heights of the pad asperities are also non-uniform (cross section of Figure 2.5 b), and they are suggested to have a lognormal height distribution [Yu94]. Therefore, at a given pressure only some portion of the pad asperities are expected to be in contact with the wafer surface and the contact is expected to increase with the increasing pressure [Ste97]. The abrasive particles are trapped and released from the pad surface continuously during polishing at the points of contact [Kir94]. However, pad surface tends to deform upon polishing and needs to be conditioned after every use. This results in wearing of the pad and changes in the pad thickness. Figures 2.6 a and b show SEM pictures of a new pad (Rodelâ€™s Suba IV) material and a glazed pad surface after polishing, respectively [Ste97]. The ability of the
19 pad to carry abrasive particles decreases as its structure degrades resulting in lower material removal rate response. Recently, fixed abrasive pads were introduced as an alternative to the regular polishing pads. In this pad structure, abrasive particles are embedded into the pad surface and polishing is conducted with abrasive free solution (Figure 2.7). The top micro-replicated resin layer is made of polycarbonate, while the sublayer is a soft foam material (vinyl acetate). The embedded abrasives are usually cerium oxide or alumina with a (a) (b) Figure 2.5. IC-1000 polishing pad surface (a) SEM micrograph (200x); (b) AFM micrograph and cross-section.
20 (a) (b) Figure 2.6. SEM micrograph of Suba IV pad surface (150x). (a) Before polishing new pad; (b) After polishing (glazed) [Ste97]. Figure 2.7. SEM micrograph of fixed abrasive pad surface (200x) [Vel00]. mean size of 0.2 micrometer [Vel00]. The use of fixed abrasive pads in CMP eliminates the colloidal stability problems allowing the use of wider variety of solution chemistry. The problems of uniform slurry distribution on wafer surface and the variations in polishing due to pad conditioning are also minimized [Ngu01]. In addition, post CMP cleaning and effluent filtration for environmental protection become more manageable
21 due to the absence of particles in the slurry [Fle98]. But the main advantage of fixed abrasive pads in processing was observed in copper polishing, where dishing and erosion problems were encountered. It was shown that the planarization was almost three times faster than the regular CMP processes and the polishing uniformity also improved with fixed abrasive pads in copper polishing [Vel00]. In another study, it was reported that the slurry free copper CMP system had lower sensitivity to overpolish time, which helped in reducing dishing and erosion [Tri99]. However, the fixed abrasive polishing was observed to create more scratches on the polished surfaces and therefore requires more improvements before it is implemented [Bra00]. Polishing Slurry CMP slurries provide both the chemical action through the solution chemistry and the mechanical action through the abrasive particles during polishing. High polishing rates, planarity, selectivity, uniformity, post CMP cleaning efficiency (including environmental concerns), shelf-life, and dispersion ability are the factors which need to be considered to optimize the slurry performance [Ste97]. Abrasives in the slurries play the very important role of transferring mechanical energy to the surface to be polished. The most commonly used abrasives in CMP are SiO 2 and Al 2 O 3 , although CeO 2 is also used for glass polishing. Even though mechanical grinding by the abrasive particles can provide planarization, this is not desirable because of extensive surface damage after polishing. The chemistry alone, on the other hand, cannot achieve planarization since most of the chemical reactions are isotropic [Ste97]. In CMP, the synergy between the chemical and mechanical interactions is utilized to achieve planarization. Chemical components of the polishing slurries alter the properties of the top surface layer of the substrate, while the abrasive particles interact with this
22 layer under the applied pressure to remove it. Successful polishing is believed to take place when the rates of chemical and mechanical activities are comparable in polishing operation [Ste99]. Therefore, it is important to determine the extent of the chemical activity on the wafer surface and the conditions under which the mechanical material removal takes place. To address these issues, in the following sections, the impact of slurry chemistry and slurry particulate properties on polishing performance is discussed. Effect of slurry chemistry on CMP performance Metal CMP In metal CMP, slurry chemistry is modified to alter the properties of the metal surface to be polished. Formation of chemically modified layer is required to provide high material removal rate, acceptable surface quality and successful planarization in metal CMP. In tungsten polishing, formation of tungsten oxide on the tungsten metal layer has been shown to be necessary to achieve satisfactory material removal [Bie98]. As can be seen in Figure 2.8, negligible material removal was obtained in the absence of the oxidizer (K 3 Fe(CN) 6 ) in tungsten CMP [Bie98]. Furthermore, the chemically altered layer must be a passive oxide layer to enable topographic selectivity during polishing [Ste94]. Figure 2.9 illustrates the proposed polishing mechanism of tungsten, which leads to the planarization of the wafer surface by achieving topographic selectivity [Kau91]. According to this mechanism, the passive oxide layer on the tungsten wafer is removed by the abrasive particles at the higher levels, while the metal at the recessed area is protected from further etching by the formation of passive film. In appropriate chemistries, formation of protective oxide layer was demonstrated by wear experiments for tungsten polishing by rubbing the surface using pin-on-disk method. As illustrated in Figure 2.10, formation of the passive oxide layer on the surface at motionless stage
23 Figure 2.8. Tungsten polishing in the absence and presence of oxidizer (K 3 Fe(CN) 6 ) [Bie98]. decreased the current to near zero, whereas when the fresh metal surface was exposed to the slurry by rubbing, current increased as the oxidation reaction started on the bare tungsten surface [Bie98]. These findings indicated that the continuous formation and removal of the passivated layer enables the material removal and planarity for metal CMP. To achieve an acceptable surface quality, the mechanical properties of the chemically activated films need to be studied. It was shown by nanoindentation analysis on tungsten and copper films (oxidized under CMP conditions) that the chemically modified surface films were harder than the metal itself [Ram00]. The forces required to remove the relatively harder metal oxide layers by plastic deformation can also cause
24 Figure 2.9. Mechanism of tungsten polishing and planarization in the presence of passive oxide layer [Kau91]. deformations on the softer metal surfaces. Hence, it is necessary to propose alternative material removal mechanisms for the chemically altered layers other than plastic deformation, to obtain good planarity and low defect counts after CMP. It was
25 Figure 2.10. Current recorded during wear experiments of tungsten for anodic applied potential (+2V) in 0.5 M H 2 SO 4 for a sphere motion cycle of 0.2 seconds [Bie98]. mentioned earlier that the chemically modified films must be protective for an effective metal CMP. The protective oxide film must be continuous, pore free and adherent. The tendency of the oxide film to protect the metal from further oxidation is related to the relative specific volumes of the metal oxide and metal. When an oxide film forms at the metal/oxide interface, the volume change due to the formation of oxide can be expressed by the Pilling-Bedworth ratio [Xu00], 00 MMAARatioPB (2.4) where, A o is the molecular or formula weight of the oxide, A M is the atomic weight of the metal, and O and M are the oxide and metal densities, respectively. It is generally
26 accepted that when the P-B ratio < 1, which means that the volume of the oxide is less than the metal volume (or the lattice constant, a, of the oxide film is smaller than the metal), tensile stresses develop within the oxide film. As the thickness of the oxide layer increases, the oxide film starts to crack to relieve the strain and becomes porous. Consequently, the underlying metal film maybe etched locally and cannot be planarized uniformly. On the other hand, if the volume of the oxide is much greater than the metal (P-B ratio >> 1), compressive stresses will start to develop as the film grows. The oxide film releases the strain energy by breaking the bonds at the metal-oxide interface resulting in lack of protective oxide film. An ideal protective oxide can be obtained when the P-B ratio is between 1 and 2. In this regime, the oxide formed on the metal surface remains intact. Its growth is limited with diffusion of the metal ions through the oxide film. Figure 2.11 represents the thin oxide film behavior as a function of the P-B ratio. For a given metal undergoing oxidation, the P-B ratio may provide a tool to predict the stress development in the film structure and the morphology of the oxide film. As the oxide-metal interface stress characteristics can strongly influence the surface quality upon polishing, the P-B ration criteria can be utilized to select alternative material removal mechanisms for metal CMP to achieve acceptable surface quality. The stability of the oxide film on the metal surface depends on the strength of the bonds at the metal-oxide interface. As the film grows, the stresses developed in the oxide film structure are balanced with the metal-oxide bond strength ( bond ), which can be estimated by Equation 2.5, 1/2Ic1/2SOfilmbonddKd)(E (2.5)
27 where E film is the film elastic modulus, O and S are the surface energies of the oxide film and metal substrate, K Ic is the fracture toughness of the oxide-metal interface and d is the film thickness [Coh90]. When a protective oxide film is formed, the internal stresses ( nternal ) induced by the oxide growth do not overcome the metal-oxide bond strength. But the oxide structure is under compression since P-B ratio is between 1 and 2. In CMP applications, the external stresses ( external ) applied on the oxide film by the abrasive particles should also be taken into account to evaluate the oxide film stability. The stress balance in the metal-oxide interface can be given as in Equation 2.6. bond = nternal + external (2.6) Figure 2.11. Thin oxide film behavior as a function of the P-B ratio (a) Porous oxide (P-B Ratio < 1); (b) Protective oxide (1< P-B Ratio< 2); (c) Film fracture (P-B > 2) [Zha00].
28 When the total magnitude of the internal and external stresses exceeds the bond strength, the oxide film can be removed by breaking the metal-oxide bonds at the interface. Therefore, the optimal load per abrasive particle, creating the external pressure at the metal-oxide interface, can be estimated by measuring the internal stress at the interface and comparing it to the critical stress for the bond failure ( bond ). This mechanism of material removal is expected to yield an atomically planarized surface due to the breakage of metal-oxide bonds at the interface. On the other hand, if the formed oxide is very brittle it may fracture under the applied pressure of the abrasive particles before the bond strength at the interface is exceeded resulting in local etching and inefficient planarization. This suggests that the mechanical deformation of the chemically altered layer (oxide layer) is possible under very low loads (below the yield strength of the metal) and the harder oxide layer can be removed by protecting the much softer metal surface underneath. Either the brittle fracture or the peeling of the oxide film can be alternatives to plastic deformation in metal CMP applications for the mechanical removal of the passivated top films without damaging the wafer surface. By determining the mechanical properties of the metal oxide layer, the mode of material removal can be predicted. Therefore, effective slurry designs can be developed including the chemistries to form a protective oxide layer and particulate properties (particle size and solids concentration) to apply adequate pressure on the oxide film to result in sufficient material removal with minimal defect formation. Silica CMP Silica is the most commonly used oxide in IC manufacturing as a dielectric material [Ste97]. In silica CMP, formation of a chemically modified layer has been
29 suggested to be necessary to achieve high material removal rate with acceptable surface quality [Izu79]. The chemical modification of the surface in silica CMP is proposed to be provided by the water molecules [Tom94, Coo90]. Indeed, polishing rates were observed to be near zero in hydrocarbon liquids such as kerosene [Izu84], paraffin [Cor63] or oil [Izu79] as well as in liquids without hydroxyl groups, such as formamide [Cor63]. Surface quality was also detected to be poor in such liquids. Therefore, it was proposed that the interactions between the siloxane bonds (Siâ€”Oâ€”Si) and water, primarily determine the behavior of silica surface during polishing. The reaction for the breakage of the network forming siloxane bonds with water to form a hydrated surface is as follows [Ile79]. H 2 O + Siâ€”Oâ€”Si Siâ€”OH + OHâ€”Si Budd explained this reaction as the attack of the solvated hydrogen ion (H 3 O + ) on the negative oxygen sites of silica [Bud61]. The reaction is completed in three stages. Initially, a water molecule from the environment attaches to the Siâ€”Oâ€”Si bond by aligning itself due to the formation of a hydrogen bond between its hydrogen and the oxygen atom of the silica surface, and interaction of the lone pair orbitals of the oxygen in water structure with the Si atom. In the second stage, a reaction occurs in which proton transfer from water hydrogen to the oxygen of silica is accomplished simultaneously with electron transfer from the oxygen of water to the silica atom. As a result of these reactions, two new bonds are formed by destroying the original bond between the silica and oxygen resulting in Siâ€”Oâ€”H bonds on the surface. The rate of this reaction is believed to be controlled by the diffusion of water into silica structure [Dor73]. The diffusion coefficient of water in silica at ambient conditions and neutral pH
30 values were measured to be quite low [Nog84, Lan85]. A significant increase in diffusion rate of water into silica structure occurs at high pH values (above pH 9.0). Iler reported three orders of magnitude increase in silica dissolution rate as the pH value was increased from 2 to 11 [Ile79]. This behavior is explained based on the OH ion acting as a catalyst for attack by water on the silica network [Bud61]. The rate of attack is much faster with OH ions than it is by water, since they create excess of electrons resulting in higher negative surface potential and consequently more attack by H 3 O + . The increase in pressure was observed to be the other factor increasing water dissolution. Ito and Tomozowa have shown that dissolution of silica increased significantly as a function of increased pressure from 50 to 250 MPa [Ito81]. Further investigations revealed that the tensile stresses were responsible for the increased silica dissolution, suggesting that the tension created on the silica surface by the abrasive particles as they trail on the wafer resulted in enhanced material removal [Nog84]. In agreement with these observations, silica polishing was observed to increase significantly above pH 9 [Izu84]. However, the polishing rate was observed to be very slow without the abrasives, which can be attributed to the ineffective transfer of the applied pressure to the wafer surface in the absence of particles [Izu79]. The transmission electron microscopy (TEM) analyses combined with the Fourier-transform infrared spectroscopy (FTIR) have revealed evidence of chemical/structural modification of the silica surface after CMP [Tro94]. The cross sectional TEM micrographs of the silica surface after CMP indicated two subsurface regions, which were described as a 2 nm surface region possessing lower density than the bulk and underneath, a 15-20 nm region with a higher density according to the grazing
31 angle X-ray density measurements. FTIR analysis confirmed that these were hydrated layers. There is also indirect evidence of the formation of a type like layer on the silica surfaces in water environment. Adler observed a short-range repulsion on silica surface by direct measurement of surface forces [Adl01]. By evaluating the water-silica system behavior under a variety of solution environments he concluded that this behavior could be explained due to the swelling of the silica surface as a result of water diffusion into structure. The gel-layer, on the surface of silica is suggested to be softer than the silica itself. This is in agreement with the structure weakening of silica by the attack of water, which was observed to result in fracture of silica surface at relatively lower stress intensities in the presence of water [Mic82]. Direct evidence was presented by Tomozava et al by measuring the hardness of silica glass as a function of water content in oil based solution [Tom94]. As illustrated in Figure 2.12, silica hardness was observed to decrease drastically with the increasing water content. The surface analysis of the indentations also showed increased plastic deformation at higher water contents, as Figure 2.12. Vickers indentation hardness of silica glass as a function of water content [Tom94].
32 shown in Figure 2.13. These observations revealed that the presence of water is not only required for enabling the material removal but also to achieve acceptable surface quality during CMP. As mentioned earlier, surface quality was observed to be poor in environments other than water, where the chemically modified layer was absent. Furthermore, at low pH slurries, where the dissolution rate of silica was minimal, surface Figure 2.13. Scanning electron micrographs of Vickers indentation of silica glass with different water contents in toluene (a) ~0 wt% water; (b) ~6.11 wt% water; (c) ~8.65 wt% water [Tom94].
33 qualities were also reported to be unacceptable [Izu79], which maybe attributed to the rate of chemically modified film formation being not sufficient. Consequently, it may be concluded that it is necessary to achieve a balance between the chemical and mechanical interactions during CMP to achieve optimal polishing performance. Since the particles are suggested to improve the dissolution reaction of silica by inducing pressure and simultaneously resulting in its removal, the particulate properties of the slurries must also be optimized for an effective oxide polishing, which is discussed in the following section. Effect of slurry particulate properties Slurry particulate properties, such as particle size, size distribution and solids concentration results in variation in the number of particles interacting with the wafer surface during polishing. This may in turn vary the load applied per abrasive particle. Therefore, slurry particulate properties must be engineered to obtain optimal polishing performance. Many researches have conducted studies to analyze the effect of particle size on material removal rate during CMP. However, the reported results are often inconsistent. Brown and Jairath showed that polishing rate increased with increasing particle size [Bro81, Jai94]. Izumitani, on the other hand, observed a decrease in material removal with increasing particle size [Izu79]. Contradicting with these findings, Cook and Sivaram concluded that the polishing rate was independent of particle size [Coo90, Siv92]. Most likely these discrepancies are primarily due to the particle size and solids loading regimes used by various researchers. To elucidate the reasons for the observed differences, tungsten and silica CMP studies were conducted over a wide range of particle sizes and solids concentrations in our group earlier [Bie98, Mah00]. In tungsten CMP, material removal rate was observed to increase with increasing solids loading and
34 decreasing particle size. Surface morphology analyses conducted on the polished samples showed no deformation on the wafer surfaces, except for a very few shallow (5 nm) scratches on the wafers polished with 10 m particles. From these findings it was suggested that in tungsten polishing, material removal mechanism was not a scratching type process but controlled more by the chemical activity enhancement by the increased contact area between the abrasives and the polished surface. On the other hand, in silica polishing, there was a continuous increase in the material removal rate as a function of the solids concentration with only 0.2 m particles. As illustrated in Figure 2.14, a decrease was observed with the 0.5 m and larger size particles after a maximum was reached. Silica wafers polished with these larger size particles showed significant Figure 2.14. Material removal rate of silica as a function of the solids loading of silica slurries for different particle sizes [Mah00].
35 surface deformations, while the 0.2 m size polishing produced a smooth surface finish. These observations of varying surface quality and material removal rate with the particle size and concentration indicated that there may be different material removal mechanisms becoming predominant as a function of the slurry particle size and concentration. To explain the dependence of material removal on slurry particle size and concentration, two polishing mechanisms were proposed by Singh et al [Bie98]. These are the contact area and the indentation based mechanisms. The mathematical expressions relating the material removal rate to the solids loading and particle size in these relationships were derived from Brownâ€™s [Bro81] expression for penetration depth of an abrasive particle onto a metal surface. Figures 2.15-a and b schematically represent the contact area and indentation based mechanisms, respectively. According to the contact area based mechanism, material removal rate is proportional to the total area of contact between the abrasive particles and the wafer surface. This mechanism was suggested to enhance the chemical activity resulting in surface modification as the particles touch the wafer surface [Bie98]. According to our further analysis, we believe that in contact area mechanism the abrasive particles are in interaction with the chemically modified layer. They help its removal by the alternative Figure 2.15. Schematic representation of material removal mechanisms. (a) Contact area mechanism; (b) Indentation volume based mechanisms.
36 mechanisms suggested in the previous section rather than the plastic deformation, which may result in deformations on the underlying substrate. Consequently, a better surface quality is expected when polishing occurs predominantly with contact area based mechanism. The total area of contact, A, increases with increasing particle concentration at a given particle size, and decreasing particle size at a fixed concentration. Therefore, it is directly related to solids concentration, C o , and inversely related to the particle size as expressed in Equation 2.7 below. This relationship predicts that higher material removal rates should be achieved with smaller particle size and higher solids loading slurries as it was observed experimentally on tungsten CMP [Bie98]. Consequently, this mechanism is expected to be predominant when small size abrasive particles are used at high solids concentrations. 31310CA (2.7) The indentation-based mechanism explains the material removal by the indents of the abrasive particles into the wafer surface. In this mechanism, material removal is suggested to take place more as a result of the mechanical interactions of the particles creating abrasions on the surface to be polished. The total indent volume, V, is inversely proportional to the concentration of particles and directly proportional to the particle size as per Equation 2.8 below. The particle indent increases as the pressure per particle increases, which is possible by decreasing the number concentration of particles on the surface at a given download. Hence, the total indent volume is higher at lower particle concentrations and when larger particles are used at a fixed concentration. This mechanism becomes predominant at low solids concentrations or with larger size particles.
37 34310C V (2.8) These two mechanisms may also explain the formation of scratches and pitting during CMP when the polishing slurries show a non-uniform size distribution, which may lead to a variation in the applied load per particle. The larger size particles may be exposed to higher loads resulting in their indentation to the wafer, while the smaller sizes remain in contact with the substrate. Especially, presence of a few larger size particles, which lead to an increase in the coarse tail end of the size distribution, is suspected to be the main reason for the surface deformations in CMP [Coo90, Sch95]. Therefore, Cook suggested the use of monosized slurries for polishing. Furthermore, slurry particle size distribution and stability must be monitored to prevent large deviations in the particle size distributions during CMP. The contact area and total indent volume based relationships are the first efforts in relating the particle size and concentration to the material removal rate response. However, it is necessary to expand the given relationships into a model to be able to predict the overall polishing performance. The modeling approaches for the CMP applications are reviewed in the following section. Modeling Efforts in CMP The earliest and most cited model used to predict material removal rate in CMP applications is that of Preston, which was developed for glass polishing [Pre27]. According to the Preston equation, material removal rate (thickness of the material removed dH, in a unit time dt) is directly related to the process variables, local pressure P and relative velocity V, as given in Equation 2.9. PVKdtdHP (2.9)
38 Other variables that may affect the polishing performance such as the abrasive properties, slurry chemistry and the tribological interactions at the wafer-slurry-pad interface are assumed to be accounted for in the Preston coefficient K p . This model is an empirical way of relating the operational variables to the material removal rate for a system where the K p value is determined by experimental measurements. However, the effects of chemical interactions on the properties of the wafer surface, or the impacts of variations in slurry particle size on polishing performance cannot be predicted based on this model. Brown [Bro81] approached polishing of metals at a single particle-surface interaction level and modeled the penetration of a single abrasive particle into metal surface using Hertzian elastic interaction theory. Equation 2.10 represents Brownâ€™s model, which is a modification of the Preston equation. The coefficient K p is replaced with 1/2E to account for the material properties of the surface to be polished, where E is the Young's modulus of the polished metal. His studies on optical polishing of metals showed that this model gave reasonable approximation in estimating the material removal rate for some materials such as nickel. PV21dtdHE (2.10) Cook tried to validate Brownâ€™s model for glass polishing and observed that the model overestimated the polishing rates by over an order of magnitude [Coo90]. To explain this discrepancy he proposed that the material properties of the surface could be different from the bulk material. Formation of such a layer may change the mechanical properties of the surface to be polished significantly. The third modeling approach is based on hydrodynamics in which the slurry is treated as a Newtonian fluid and the pad and wafer are treated as rigid surfaces. The
39 most cited hydrodynamic model was introduced by Runnels [Run94a]. In this model he assumed the existence of a hydrodynamic boundary layer between the pad and wafer which led to material removal [Run94b]. According to Runnels approach, the material removal rate equation is represented by, nPCdtdH (2.11) where C p is a Preston-like constant, is the normal stress tensor due to the applied force and is the shear stress due to slurry flow activated by relative motion between pad and wafer. Similar assumptions were also used by Agarwal and co-workers to determine the stress distribution on the wafer surface with two-dimensional and three-dimensional features using the finite element method [Aga00]. Their simulations indicated that it maybe possible to remove features on wafers by hydrodynamic forces alone. However, it is known that the polishing pad is not a rigid surface as assumed in these models. Furthermore, these investigations did not present any experimental data in support of the proposed predictive methodologies. Therefore, the validity of the hydrodynamic models and their extent of accuracy are not known. Tseng and Wang combined Equation 2.11 and the Hertzian particulate penetration approximation to obtain an alternative removal rate equation expressed as: 2165VMPdtdH (2.12) where M is a constant dependent on process conditions, and P and V are the head pressure and velocity, respectively [Tse97]. This model was extended by Shi et al, by accounting for the polishing pad properties [Shi98]. The resulting removal rate
40 expression was represented as in Equation 2.13. However, this approach was also limited to the simulation of specific polishing systems. VPKdtdH32P (2.13) It can be seen from the above descriptions that the usual trend in CMP modeling is to modify the Preston equation to improve the agreement between material removal rate response and the experimental results. There are other studies in which, detailed modifications of the Preston equation were introduced for copper [Luo98, Mau97] and silica CMP applications [Oum99]. Although these models are adequate for specific applications, they concentrate only on the operational variables and are not applicable to other systems with different slurry properties. In addition, chemically altered layer formation is not taken into account, which is important to predict surface quality as well as the material removal rate in CMP. Evidently, there is a need for developing more fundamental predictive methodologies to control the CMP processes, particularly to keep up with the smaller size device manufacturing. Scope of the Dissertation This study concentrates on the impact of slurry particle size distribution and stability on the CMP performance by analyzing the possible changes in the pad-particle-substrate interactions during polishing. Furthermore, methodologies are developed to formulate stable slurries that will lead to the controlled material removal and optimal surface quality with a potential application for improved slurry selectivity. Finally, the impacts of particulate properties on polishing performance with possible polishing mechanisms are investigated. The silica-silica polishing is studied in this work as a model system. It is expected that the reported findings can be extended to the other types
41 of CMP applications. Formulation criteria for engineered particulate systems introduced in this dissertation is also believed to assist in development of better predictive methodologies for the CMP processes.
CHAPTER 3 DETECTION AND IMPACT OF HARD AGGLOMERATES IN CMP Introduction The rapid advances in the microelectronics industry require a decrease in the size of microelectronic devices [Lea97]. Fabrication of these small feature devices without defects necessitates significant improvements in the CMP process. The success of CMP operations depends on the rate of material removal and the quality of the surface finish. Optimal removal rate and smooth surface finish are achieved by the synergistic effects of the chemical and mechanical forces encountered in the CMP process. It has been demonstrated that in oxide CMP, the chemical content of the polishing slurry is responsible for enabling erosion, whereas mechanical forces help to achieve the required planarization and uniformity [Run96]. The mechanical action in CMP is mostly provided by the submicrometer size abrasive particles as they flow in between the pad and the wafer surface under the applied pressure. Cook reported that the surface damage was less when monosized slurries were used for polishing [Coo90]. However, in practical applications slurries may exhibit a size distribution due to contamination or particle agglomeration. A small fraction of hard, coarse size particles (hard agglomerates) in these slurries suggested to be responsible for deformation on the wafer surfaces, resulting in defective microprocessors [Sch95]. Hence, the detection and removal of oversized particles at small concentrations is critical for efficient CMP operations. Structural damage on wafer surfaces can be classified as scratches, pits, delamination of film interfaces and introduction of chemical or particulate impurities 42
43 [Ste97]. Among these, the major defects created by oversize particles are the formation of microscratches and pitting of the wafer [Sch95]. The importance of detecting the coarse fraction in CMP slurries has recently received added attention, as the reduction of surface defects has become more important due to the continuously decreasing size of microprocessors. It has been shown that the static laser light scattering technique is capable of detecting the 1.0 m particles in 0.4 m size baseline slurry, and can be used to monitor the removal of the unwanted coarser size abrasive particles from the slurry [Poh96]. The impact of the slurry particle distribution on defects and the vulnerability of the post CMP cleaning process was studied by Nagahara et al using Accusizer 770 [Nag96]. This instrument was found to be suitable for detecting the particles greater than one micron size. A direct correlation was reported between the amount of the large slurry particles and the particles remaining on the wafer. In another study, the sensitivity of acoustic spectroscopy technique was investigated by Dukhin and Goetz for detecting the coarser particles in a fine particle size slurry [Duk98]. They reported that the detection limit can be as low as a single 1.0 m particle per 100,000 particles of 0.1m size. However, no quantitative results were presented to establish the impact of coarser particles on polishing performance at the established detection limits. In this chapter, results of a study on establishing the limits of a static light scattering technique in detecting a small number of larger particles in a commercial CMP slurry are presented. Simultaneously, the chemical mechanical polishing responses of the wafers polished with the slurries containing hard agglomerates at the established detection limits are discussed in terms of surface topography changes, defect formation and material removal rate responses.
44 Experimental Particle size analyses were conducted by using Coulter LS 230 static light scattering instrument with a small volume module. Rodel 1200 commercial fumed silica slurry with a mean particle size of 0.14 m was used as the baseline CMP slurry (supplied by Rodel Inc.). This slurry was spiked with sol-gel silica particles of 0.5, 1.0 and 1.5 m sizes that were obtained from Geltech Corporation. To establish the detection limits of light scattering technique, 20 ml slurries of Geltech silica particles were prepared at 5 wt% solids at pH 10.5 using the Rodel 1200 slurry supernatant to preserve the original chemical composition of the baseline slurry. The baseline Rodel 1200 slurry was also diluted from 12 wt% to 5wt % solids concentration using its own supernatant. The mean particle size of the baseline slurry was found to be the same after dilution. The Geltech sol-gel silica slurries were sonicated for 40 to 60 minutes in an ulrasonic bath until all the agglomerates were broken and the mean particle size matched with the particle sizes observed by SEM analysis. The coarser size Geltech particles were mixed with the baseline slurry using a microliter syringe at a total volume of 500 microliter. The spiked slurries were sonicated for an additional 15 minutes and then fed into the analyzer. The measurements were conducted by using pH adjusted water (pH=10.5) in Coulter LS 230 to prevent the agglomeration of particles during the measurements. The polishing tests were performed on p-type silicon wafers on which a 2 m thick SiO 2 layer had been plasma enhanced chemical vapor deposited (PECVD) (supplied by Silicon Quest International). The 8-inch wafers were cut to square samples of 1.0 x 1.0 inch square and Struers Rotopol 31 tabletop polisher was used for polishing with IC
45 1000/Suba IV stacked pads supplied by Rodel Inc. The Struers Rotopol 31 tabletop polisher has the advantage of studying the polishing performance of a selected system in detail with minimal expense. This system is capable of simulating the real CMP applications and determining the material removal rate, surface quality, planarization and uniformity responses. A grid-abrade diamond pad conditioner was utilized to abrade the pad before conducting each polishing test. The downforce was 7.0 psi (492 g/cm 2 ) and the rotation speed was 150 rpm both for the pad and the wafer. The thickness of the oxide film on the wafers was measured via spectroscopic ellipsometry method before and after the polishing to calculate the removal rate. The slurry flow rate was 100 ml/min. The polishing tests were conducted with 50 ml slurries for 30 seconds at a solids concentration of 12 wt%. Atomic Force Microscopy (AFM) technique was used for the surface roughness and deformation analysis of the polished wafers. A minimum of four polishing tests were conducted for all conditions and five 10 m x 10 m size images were taken on each polished wafer to evaluate the root mean square (RMS) surface roughness and maximum surface deformation (maximum depth of scratches or pits on the surface, R max ) responses. Results and Discussion Detection of Coarser Particles in CMP Slurries The limits of static light scattering technique in detecting a small amount of coarse particles in CMP slurries were established using two different methods. The first method was based on detecting a shift in the mean diameter and size distribution curve of the baseline slurry, as it was gradually spiked with increasing concentration of the larger size particles. The mean size of the spiked slurries increased with increasing number of the coarser particles. However, until a critical concentration of coarser sizes was added
46 in the baseline slurry, the change in mean size was not statistically significant. When the critical concentration was reached, both mean diameter and size distribution curve of the spiked slurries were observed to shift significantly. Therefore, this particular concentration of the coarser particles was set as the â€œcurve shift detection limitâ€. The second method was based on the appearance of a second peak at the size range of the spiked particles. Above a certain concentration of the spiked coarser particles, the Coulter LS 230 software started to fit the data to a multimodel distribution, representing the coarser particles as an individual peak in addition to the peaks obtained with the original slurry. The concentration of larger size particles required to form an extra peak was defined to be the â€œdouble peak detection limitâ€. Figure 3.1. Particle size distributions for Rodel 1200 slurry spiked with Geltech 1.5m particles [Bas00].
47 Figure 3.1 shows the change in the particle size distribution of Rodel 1200 slurry as a function of Geltech 1.5 m coarse particle addition. The mean particle size of the Rodel slurry was 0.14 m and it originally exhibited a bimodal distribution with the main peak at 0.1 m and a second peak at 0.5 m. The analysis conducted on the Rodel slurry by using Accusizer 780, which is a number counting technique, also showed the presence of the larger particles. It is suspected that the fused silica chains in the Rodel slurry entangle in non-spherical shapes and detected as larger particles due to the scattering through the longer dimension. As the 1.5 m Geltech particles were gradually added into the baseline Rodel slurry, a statistically significant shift of the mean particle size was observed at 0.2 wt% addition of the coarser fraction. When the amount of coarser particles was increased further, an individual peak was detected at 1.5 m size range at a Table 3.1. Detection limits of Coulter LS 230 for Rodel 1200 slurry spiked with 0.5, 1.0 and 1.5 m Geltech particles.
48 concentration of 1.1 wt% representing the double peak detection limit. Table 3.1 summarizes the detection limits of Coulter LS 230 for the Rodel slurries spiked with 0.5, 1.0 and 1.5 m Geltech particles. Since the original Rodel slurry had exhibited a peak at 0.5 m range, the second peak detection could not be established for the slurries spiked with 0.5 m particles. It was observed that the sensitivity of coarse particle detection improved as the difference in the mean particle size of the baseline slurry and the spiked coarser particle size increased. In addition, the curve shift method was able to detect the larger particles at relatively lower concentrations. It was found that, one 1.5 m particle for every 460,000 Rodel particles (0.14 m) can be detected using the curve shift detection limit. This ratio was 1:84,000 using double peak detection protocol. However, the curve shift method can be used only if there is information available about the characteristics of the baseline slurry (such as mean particle size and standard deviation of the distribution). Surface Roughness and Critical Defect Analysis on the Wafer Surfaces in the Presence of Coarser Particles During CMP To examine the surface morphology and defect formation in the presence of coarser particles, polishing tests were conducted with the slurries prepared by mixing Geltech coarse particles into the baseline Rodel slurry at the established detection limits of the static light scattering technique. Subsequently the polished wafer surfaces were analyzed with atomic force microscopy. Figure 3.2 shows the three-dimensional AFM pictures (5 m x 5 m) of the surfaces polished with the baseline Rodel 1200 slurry and the slurries spiked with 1.1 wt% of 0.5 m (statistical shift detection limit) and 1.5 m (double peak detection limit) Geltech particles. It is clearly seen in Figure 3.2 a that polishing with Rodel slurry
49 yielded acceptable polishing results. The maximum depth of surface defects (R max ) was less than 3nm and the RMS roughness was 0.5 nm. However, the slurry spiked with 1.1 wt% coarser particles, resulted in surface deformation, which increased with the increasing particle size. Pitting and scratch formation was observed on the oxide film as Figure 3.2. AFM pictures and cross sections for the wafers polished with Rodel 1200 slurry before and after spiking with coarser size particles. (a) Rodel 1200 baseline slurry; (b) Rodel 1200 slurry spiked with 1.1 wt% Geltech 0.5 m particles; (c) Rodel 1200 slurry spiked with 1.1 wt% Geltech 1.5 m particles [Bas00].
50 seen on Figures 3.2 b and 3.2 c. R max values reached 15 nm and 40 nm with 0.5 and 1.5 m size particles, respectively. The formation of these defects may be explained by uneven distribution of the applied load on the abrasives due to the variation in size. The larger particles are expected to carry more load than the smaller particles resulting their indentation on the wafer surface leading to defects. Figure 3.3 shows the surface and cross-section analyses of the wafers polished with Rodel slurries spiked with 1.0 m particles at curve shift (0.3wt%) and double peak (a) (b) Figure 3.3. Cross-section analysis for the wafer polished with spiked Rodel 1200 slurry. (a) Slurry spiked with 0.3 wt % Geltech 1.0 m particles (statistical shift detection limit); (b) Slurry spiked with 1.3 wt% Geltech 1.0 m particles (double peak detection limit) [Bas00].
51 (1.3 wt%) detection limits. The effects of increasing coarse particle concentration are evident in these pictures. At 0.3 wt% concentration of 1.0 m particles (Figure 3.3 a), the surface was relatively smooth but scratch formation was detected. This can be attributed to the relatively higher pressure per particle at low concentrations (0.3 wt%) of coarse particles compared to the higher concentrations (1.3 wt%). Due to the higher pressure applied on the particles, they tend to be embedded into the pad and the silica surface resulting in scratches as they slide on the wafer. The calculations showed that the width of the detected scratch matched the dimensions of one that could be formed by a 1.0 m particle at the measured depth. As the concentration of 1.0 m particles increased to 1.3 wt% (Figure 3.3 b), the surface roughness increased and pitting type of deformations started. It is suggested that as the pressure per particle is reduced by the increasing number concentration of the coarser particles, the abrasives perhaps started rolling on the wafer surface rather than sliding, as it will be discussed in detail in Chapter 6. Table 3.2 summarizes the quantitative results of the polishing tests. It is clear that the surface roughness and the depth of surface defects (R max ) increased with the increasing size and concentration of coarser particles in the spiked slurries. The standard deviations of RMS roughness between similar samples were also greater at higher concentration of coarser particles, indicating higher overall surface damage. In general, the presence of the coarser particles resulted in a significant increase in the surface roughness and surface defects on the wafer surfaces. Material Removal Rate Response in the Presence of Coarser Particles The removal rates obtained with and without spiking the Rodel slurry with the larger particles are presented in Figure 3.4. The oxide removal rate of the baseline Rodel1200
52 Table 3.2. Summary of the polishing results for the Rodel 1200 slurry spiked with 0.5, 1.0 and 1.5 m Geltech silica particles at the established detection limits of Coulter LS 230. slurry was measured to be 10,000 /min. As the coarser particles were spiked into the Rodel slurry, the removal rate decreased, except for the slurries spiked with 1.5 m particles at 1.1 wt% concentration for which the removal rate was 10,300 /min. Table 3.2 also summarizes the material removal (polishing) rates. The variation in the material removal rates can be explained based on the changes in particle-substrate interactions in the presence of hard agglomerates. In the previous chapter, dependence of material removal on the total abrasive contact area and indentation volume was explained in detail based on two mechanisms [Bie98]. Briefly, the total contact area mechanism is thought to enhance the chemical activity on the surface due to the interactions between the abrasives and film [Izu79, Nog84, Tro94, Tom94]. The total area of contact will increase with the increasing particulate concentration and decreasing particle size. The indentation-based mechanism, on the
53 Figure 3.4. Polishing rates for the baseline and spiked Rodel 1200 slurries [Bas00]. other hand, explains the material removal via particle indentations. This creates more mechanical removal and the total indent volume increases as the pressure per particle increases by decreasing the number concentration of particles or increasing the particle size. Overall polishing is believed to be governed by a combination of these two mechanisms. Figure 3.5 schematically depicts the proposed mechanism of material removal for the slurries spiked with the coarser particles based on the interplay between the contact area and indentation mechanisms. It is suggested that the coarser particles act like pillars and tend to hold the wafer away from the pad surface reducing the interaction of smaller
54 size Rodel particles with the substrate. Therefore, although the larger size particles tend to indent the surface more due to the higher load applied, the material removal rate decreases due to the decreased total contact area by the reduced number of effective abrasive particles contacting the surface. As given in Table 3.2, the depth of surface defects increases with the increasing size of the spiked hard particles perhaps due to the higher pressure per particle as the size of the hard agglomerates increased. However, at a particular population of larger particles, as observed with 1.5 m particles at 1.1 wt% concentration, the material removal rate approaches the baseline polishing rate suggesting Figure 3.5. Change in polishing mechanism for the Rodel slurries spiked with Geltech coarser particles [Bas00].
55 that sufficient contact area combined with particle indentation was reached. Yet, the critical surface damage occurring under these conditions (as seen on Figure 3.2 c) indicates that this is not an optimal size distribution for the CMP slurries. Summary The limits of light scattering technique in detecting a small number of coarser particles in a commercial CMP slurry were established using the Coulter LS 230 particle size analyzer. The results showed that the detection limits of light scattering technique could be as low as parts per million level (two 1.5 m particles for a million 0.14 m Rodel 1200 particles) using the statistical shift method. The double peak detection method on the other hand, was able to detect the coarser particles (5 to 10 times larger than the base slurry particles) at approximately 1 wt% concentration of the total solids in the slurry (one 1.5 m and five 1.0 m particles for one hundred thousand of Rodel 1200 particles). The results of the polishing tests showed that the presence of coarser particles tend to create critical defects on the oxide film and change the polishing mechanism. The surface damage increased with the increased size and concentration of the coarser particles added to the baseline slurry. It was also observed that the polishing mechanism may vary based on the size and the concentration of the coarser particles. These findings are in agreement with the previous studies. Therefore, there is a critical need to detect the coarser particles at much lower concentrations, and their removal from the CMP slurries is necessary for obtaining optimal polishing results. The coarser particles in the slurries may also present due to the agglomeration of the primary slurry particles during polishing. In the following chapter, the impact of
56 transient (soft) agglomerates on polishing performance is investigated to obtain a complete understanding on the reasons for defect formation during CMP operations.
CHAPTER 4 FORMATION AND IMPACT OF SOFT AGGLOMERATES IN CMP Introduction In Chapter 3 it was demonstrated that the contamination of a commercial CMP slurry with 1 wt% larger size spherical particles (hard agglomerates) resulted in increased surface roughness and higher number of surface defects [Bas00]. Their presence also changed the material removal rate response. To remove coarser particles, filtration of CMP slurries is commonly practiced. Nevertheless, even after filtering the slurries, the defect counts on the polished surfaces were observed to be higher than desired [Ewa99]. It has been speculated that some of the defects are caused by the transient soft agglomerate formation (agglomeration of the primary size abrasive particles) in the CMP slurries. Indeed, it was reported previously that the commercial CMP slurries tend to coagulate and partially disperse during the polishing process, confirming the presence of transient agglomerates [Cab00]. In this part of the dissertation, a systematic study was conducted on silica-silica polishing to investigate the effects of transient (soft) agglomerates on the polishing performance. To generate soft agglomerates, a part of the 0.2 m monosize silica slurry (at pH 10.5) was substituted with dry aggregated, polymer flocculated and salt coagulated fractions. The performances of the slurries were evaluated based on the material removal rate and surface quality analyses conducted on the polished wafers. Simultaneously, the changes in slurry particle size distribution during polishing were monitored. 57
58 Experimental To study the role of soft agglomerates on polishing efficiency, 0.2 m monosize sol-gel silica powder obtained from Geltech Corporation was used to prepare the baseline slurry. The pH was adjusted to 10.5 and the solids concentration was kept at 12 wt% for all the slurries. To stabilize the baseline slurry, dry silica powder was ultrasonicated in deionized water while maintaining the pH at 10.5 with NaOH addition. Preparation of the slurries from a dry powder was necessary to enable the control of system chemistry while making modifications to generate some of the aggregates. Three techniques were used to prepare slurries containing soft agglomerates. Initially, dry aggregates were left partially undispersed in the baseline slurries by controlling the ultrasonication during the slurry preparation. The dry aggregates may form during mixing of the slurries prior to polishing. In the second method, well-dispersed baseline silica slurry was flocculated using 0.5 mg/g, 8,000,000 molecular weight polyethylene oxide (PEO) polymer. To obtain homogeneously flocculated slurries, the slurry samples were stirred gently in 60-ml bottles and size analyses were conducted as a function of time. In the beginning, slurries were found to be highly flocculated and inhomogeneous. After 30 minutes of stirring, more homogeneous slurries were obtained with a main peak at 0.2 m and a second peak at 5-10 m size range. As the stirring continued, a monosized well-dispersed slurry was achieved after two hours of agitation and its size distribution curve overlapped with the baseline slurry. It was reported in the past investigations that 8,000,000 MW PEO adsorbs onto sol-gel silica surface at pH 9.5 [Mat96]. Therefore, it is suggested that initially the abrasive silica particles were flocculated by the bridging mechanism, followed by breaking
59 Figure 4.1. Particle size distribution of the baseline and soft agglomerated slurries. The soft agglomerates were prepared by dry aggregation, polymer flocculation and salt coagulation methods at 5-10 m size range [Bas02]. polymer bridges during the stirring action. Finally, in the third method, NaCl salt was added to the slurries at 0.2, 0.4 and 0.6 molar concentrations to form coagulates. Particle size analyses of the slurries were conducted by Coulter LS 230 instrument utilizing light scattering technique with small volume module. The background water used to run the size analysis was prepared to have the same chemical composition as the modified slurries. Figure 4.1 shows the size distribution curves of the baseline, dry aggregated, polymer flocculated and salt coagulated slurries as a function of differential volume. It can be seen that the baseline polishing slurry had a narrow, unimodal particle size distribution with a mean size of 0.2 m. Slurries containing soft agglomerates, on
60 Table 4.1. Slurry performance summary in the presence of soft agglomerates. the other hand, exhibited bimodal particle size distributions with a second peak at 5-10 m size range. Polishing tests were performed using the slurries containing the soft agglomerates using the procedure described in the previous chapter. Results and Discussion Effect of Dry Aggregates Table 4.1 summarizes the performance of partially dispersed/agglomerated slurries in terms of material removal rate, surface roughness (RMS) and maximum depth of surface defects (scratches or pitting) (R max ) response. For the baseline polishing, material removal rate was 3800 /min and the surface roughness was 0.85 nm with 25
61 nm of maximum surface deformation. Figure 4.2 a shows the AFM image of the silica wafer polished with the baseline slurry. The wafer surface was smooth with minimal deformations. When dry aggregates were present in the polishing slurries, the mean particle size increased to 0.77 m from the original size of 0.2 m. In the presence of dry aggregates, material removal rate was 4300 /min indicating a trend towards increased removal. The surface quality, on the other hand, degraded significantly as seen in Figure 4.2 b. Surface roughness value increased to 2.66 nm and the maximum surface deformation was detected to be 65 nm as given in Table 4.1. These results demonstrate the adverse effect of the dry aggregates on the polishing process. The higher magnitude of the attractive van der Waals interactions between the dry silica particles compared to the silica particles dispersed in a solution can lead to formation of relatively rigid agglomerates in the dry state [Isr92]. As a result, the aggregates of dry powders are believed to be the most rigid transient agglomerates among the ones studied in this investigation. During the CMP applications, dry aggregates are expected to sustain most of the applied load since their size is larger than the primary slurry abrasives. Therefore, they are exposed to high download during polishing that (a) Baseline (b) Dry aggregates Figure 4.2. AFM images of the silica wafers after CMP. (a) Baseline 0.2 m 12 wt% monosize silica slurry; (b) Slurry with dry aggregates [Bas02].
62 may result in their breakage. Indeed, the particle size analysis of the slurries collected from the pad after polishing with dry aggregated slurries showed a shift towards increased volume of primary size abrasives. Yet, while breaking, they may result in strong mechanical abrasion (indent) on the wafer surface due to their rigidity. It was observed in previous studies that the mechanically driven material removal created significant surface deformations during polishing similar to the damage created by the dry aggregates [Mah00]. Effect of Polymer Mediated Soft Agglomerates (Flocs) Polishing tests were conducted using the slurries containing PEO with and without the flocculated agglomerates. It can be seen in Figure 4.3 a that the flocculated slurries (mean particle size of 5.82 0.67 m) resulted in severe surface deformations during polishing with a material removal rate of 3680 /min as compared to the baseline rate of 3800 /min. The surface roughness value increased to 1.14 nm, R max was found to be 45 nm in the presence of flocs. Similar to the dry aggregation case, the flocs are believed to break during the polishing operation under the applied normal and shear (a) Flocculated with PEO (b) Dispersed with PEO Figure 4.3. AFM images of the silica wafers after CMP. (a) Slurry flocculated using 0.5 mg/g PEO; (b) Slurry dispersed in the presence of 0.5 mg/g PEO [Bas02].
63 Figure 4.4. Particle size distribution of the baseline and polymer flocculated slurries. The size distribution of the flocculated slurries shifted down after polishing. Slurries stirred long enough in the presence of PEO dispersed to the original size distribution of the baseline slurry [Bas02]. forces since the size distribution of the slurries collected from the pad after polishing shifted downwards as can be seen in Figure 4.4. Considering the flocs are made of well-dispersed particles bridged with the polymer chains, their formation should involve more flexible bonds compared to dry adhesion case. Therefore, the rigidity of the polymer flocs is believed to be much less relative to the dry aggregates and they are expected to deform between the pad and wafer surfaces more easily. This may explain the reduced surface deformation value obtained in the presence of flocs (45 nm) as compared to the dry aggregation case (65 nm).
64 Figure 4.3 b shows the AFM image of the silica wafer polished with the slurry stabilized in the presence of 0.5mg/g PEO. There were no flocs detected in this slurry and its size distribution curve overlapped with the baseline slurry as shown in Figure 4.3. The surface quality of the wafers polished with this slurry was comparable to the baseline polishing result. The material removal rate on the other hand, decreased to 3090 /min from the baseline value of 3800 /min. This result can be explained based on the possible lubrication of the particle and substrate surfaces by the polymer adsorption [Kle94]. For the slurries that are stable in the presence of PEO, each abrasive particle is expected to be surrounded by the polymer molecules. Therefore, the abrasive particles can be effectively and uniformly lubricated resulting in lower material removal rate but much better surface quality. To determine the change in the frictional force with the addition of PEO, AFM measurements were conducted by attaching a 7.3 m silica particle to the AFM tip. These measurements were performed under baseline conditions and with 5mg/g PEO solution. The frictional force (F F ) was measured between the abrasive particle attached to the AFM tip and the silica wafer surface as a function of the normal force (F N ) applied to the tip. The friction coefficient () was calculated from the slope of the curve based on Amontonâ€™s law (Equation 4.1). F F = F N (4.1) Figure 4.5 shows the AFM friction coefficient measurement in the absence and presence of PEO at a single-particle substrate interaction level. It can be seen that the addition of PEO resulted in a decrease in the friction. coefficient (0.12) as compared to the baseline value of 0.25. The observed decrease in the friction force in the presence of
65 Figure 4.5. AFM friction coefficient measurements for the baseline and PEO containing solutions [Bas02]. PEO indicates the lubrication effect of the polymer. In agreement with this finding, the boundary lubrication studies using atomic force/friction microscopy technique have shown that the coefficient of friction of the Si(100) sample lubricated with polymeric lubricants (Z-15 and Z-DOL perfluoropolyether) was lower than that of unlubricated sample [Bhu95a]. A similar trend was also reported for tungsten polishing in the presence of surfactants [Bie99]. When the alumina particles in the polishing slurries were stabilized using a mixed surfactant system (anionic and nonionic), 30 % less material removal rate was reported than without the presence of surfactants, however, yielding a much better surface quality.
66 Effect of Salt Coagulated Agglomerates To delineate the effect of salt coagulated agglomerates on polishing performance, NaCl was added to well-dispersed baseline slurries. NaCl was selected in this investigation since it was reported in the literature that Na salts could coagulate the silica suspensions [Ile79]. The critical coagulation concentration (CCC) of NaCl for 12 wt.% silica slurries at pH 10.5 was measured to be 0.25 M by conducting size analyses of the slurry as a function of salt concentration. Accordingly, three different salt concentrations were selected starting with 0.2 M, which is below CCC and 0.4 and 0.6 M concentrations above CCC. Figure 4.6 illustrates the particle size distribution analysis for the salt coagulated slurries. 0.2 M NaCl addition did not result in coagulation and the mean particle size of the polishing slurry remained to be 0.2 m. When the polishing tests were conducted at this salt concentration, the surface roughness of the polished wafer was comparable to the baseline polishing (0.86 nm) as can be seen in Figure 4.7 a. The R max value, on the other hand, increased to 50 nm from the baseline value of 25 nm. The material removal rate response also showed a significant increase in the presence of 0.2 M NaCl. A value of 4650 /min was obtained as compared to the 3800 /min for the baseline polishing. The increase in the material removal rate can be explained based on two phenomena. First, it is well known that both the silica wafer and the silica abrasive particles exhibit negative potential at 10.5 pH (-potential of mV at pH 10.5) leading to strong repulsive forces. To accomplish polishing, these repulsive forces must be overcome. As the ionic strength of the solution is increased, charges on the silica surface decreased lowering the repulsion [Isr92]. The decrease in the extent of the repulsive electrostatic forces was also
67 Figure 4.6. Particle size distribution of the baseline and salt coagulated slurries. The size distribution of the slurry with 0.2 M NaCl did not change significantly. Addition of 0.6 M NaCl coagulated the slurry. Size distribution of the 0.6 M NaCl slurry before and after polishing remained the same [Bas02]. reported between a silica particle and silica plate at increasing salt concentrations using AFM technique [Raj97]. The enhanced particle-surface interactions in the presence of salt may lead to more effective mechanical abrasion yielding higher material removal rate. Similarly, the surface of the polishing pad was also reported to exhibit a high negative potential (-potential -25mV) at high pH, suggesting that there are also repulsive forces between the pad and the wafer to be polished [Mah99a]. As the ionic strength of the solution is increased by the addition of salt, the screening of the charges enables closer contact between the pad and the wafer surface. This phenomenon was shown to
68 increase the frictional forces between the pad and the wafer surface by in-situ dynamic force measurements [Mah99a, Mah99b]. Consequently, the pad-surface interactions could also favor increased material removal in the presence of salt. The second explanation for increase in the material removal rate with salt addition relates to the changes in surface potential. In water, silica dissolves due to the attack of the negatively charged non-bridging oxygen atom by the solvated hydrogen ion (H 3 O + ) [Bud61]. As the solution pH is increased, silica dissolution rate increases since the OH ions create excess of electrons resulting in higher negative surface potential and consequently more attack by H 3 O + . The complete dissociation of surface groups is achieved, when the overall rate constant of dissolution (k + ) reaches the limiting value of 10 -8.25 mol m -2 s -1 [Vog99]. Similarly, silica dissolution has been shown to increase at high ionic strengths due to the increase in the surface charges by the salt addition. The complete dissociation of the surface sites is achieved at lower pH values in the presence of salt as compared to the salt free solutions [Hou92, Dov92]. In other words, at high ionic strengths, silica surface behaves as if it is in a higher pH environment than the actual solution pH. The increase in the maximum surface deformation with the addition of salt can also be related to the increasing frictional forces. Additionally, it is suggested that some transient agglomerates may be forming during polishing in the presence of 0.2 M NaCl. Theoretical calculations based on the constant charge assumption show that the magnitude of repulsive force barrier in the presence of 0.2 M NaCl is 167 kT as compared to 720 kT in the absence of salt. Thus, transient agglomerates may form in the presence of 0.2 M NaCl due to enhanced particle-particle interactions, when local variations occur in the particle concentration under the dynamic polishing conditions.
69 However, no such agglomerates were detected by the size measurements conducted with Coulter LS 230. As the salt concentration of the polishing slurries increased to 0.4 and 0.6 M, the slurries coagulated irreversibly. The mean particle size of the 0.4 M and 0.6 M NaCl containing slurries were measured to be 1.3 and 3.6 m, respectively. Figure 4.6 shows the size distribution of the 0.6 M NaCl containing slurry before and after polishing. Both curves are almost identical showing bimodal distributions with a first peak around 0.2 m and a second peak at 5-10 m size ranges. This observation indicated that the abrasive particles coagulate at high ionic strengths and even if they break during polishing under the applied shear and normal forces, they perhaps coagulate back as the pressure is released. AFM analysis conducted on the wafers polished with high ionic strength slurries showed significant degradation in the quality of the polished surfaces suggesting the adverse effects of the salt mediated aggregates (coagulates). Figure 4.7 b demonstrates the AFM image of the wafer polished with 0.6 M NaCl containing slurry. The surface roughness was measured to be 2.76 nm and maximum surface deformation increased to (a) 0.2 M NaCl (b) 0.6 M NaCl Figure 4.7. AFM images of the silica wafers after CMP. (a) Polished with 0.2 M NaCl containing slurry; (b) polished with 0.6 M NaCl containing slurry [Bas02].
70 120 nm. The significant increase in surface roughness is suggested to be due to the presence of stable agglomerates in the slurries at high ionic strengths. The material removal rate increased with the increase in salt concentration as summarized in Table 4.1. A similar observation was reported on silica CMP by Mahajan et al [Mah99a]. As explained previously, this phenomenon is believed to be occurring due to the closer pad-particle-substrate interactions by the reduced repulsive force barrier, and also as a result of the enhanced silica dissolution due to the faster saturation of the surface sites. These analyses indicate that salt addition can significantly enhance the material removal rate but results in major surface deformations. Summary In this chapter effects of soft agglomerates (aggregates of the primary size abrasive particles) on CMP slurry performance were reported. The surface quality of the polished wafers degraded significantly when soft agglomerates were present in the slurries. The magnitude of surface defects changed depending on the rigidity of the agglomerates. A small amount of relatively harder dry aggregates deteriorated the performance of CMP surface quite dramatically, similar to the situation with a few coarser particles present in an otherwise dispersed slurry. Presence of relatively softer flocs and coagulates also produced undesirable surface effects, though at a much higher concentration of aggregated particles. The material removal rate response, on the other hand, was observed to be controlled by the extent of the pad-particle-surface interactions that impact the magnitude of the frictional forces enabling abrasion and material removal. These observations indicate that to design optimally performing CMP slurries, interparticle and pad-particle-surface interaction forces must be controlled. In addition, it was demonstrated that not only the coarse (hard) particles, but also the soft agglomerates
71 of the primary abrasive particles must be avoided in the polishing slurries to ensure acceptable surface quality. This finding underlines the importance of developing robust dispersion schemes for the CMP slurries. Dispersion of the particulate systems under the severe environments of the CMP process will be discussed in the following chapter.
CHAPTER 5 STABILIZATION OF CMP SLURRIES BY CONTROLLING INTERACTION FORCES Introduction In the previous chapter, it has been shown that even soft agglomerates result in significant surface scratching and pitting during CMP. Transient soft agglomerates may form in the slurries due to the local variations in the abrasive concentration, ionic strength or pH as the particles are exposed to the polishing process. Therefore, it is critical to design robust dispersion schemes for the CMP slurries, which will remain fully stable under extreme pH and ionic strength as well as the dynamic processing conditions (high normal and shear forces) encountered during polishing. Most of the commonly used dispersion techniques, such as the electrostatic stabilization, inorganic dispersant addition or polymeric dispersion, may not perform adequately under the severe environments of CMP [Adl00]. Alternatively, surfactants are known to stabilize particulate suspensions at high electrolyte concentrations by forming self-assembled structures at the solid-liquid interface [Col97, Sil99, Koo99, Bre99, Eva96, Pal00]. Specifically, it was determined in our research laboratory that self-assembled structures of C 12 TAB (cetyl trimethyl ammonium bromide), a cationic surfactant, provided stability to silica suspensions at high ionic strength and extreme pH by introducing a repulsive force barrier [Adl00, Sin01]. In this chapter, stabilization of the silica CMP slurry at high pH and salt concentration is discussed by studying the particle-particle and particle-substrate interactions. 72
73 Experimental The baseline polishing slurries were prepared using 0.2 m monosize sol-gel silica powder at 12 wt% concentration as detailed in Chapter 4. To simulate the extreme CMP slurry environment, 0.6 M NaCl or 0.24 M CaCl 2 salt was added to the baseline slurry, which resulted in coagulation of the abrasive particles. Under these conditions, C n TAB (n=8, 10, 12) surfactant was added to the suspensions. Particle size analysis of the slurries was conducted using Coulter LS 230 instrument utilizing light scattering technique with small volume module. The background water used to run the size analysis was prepared to have the same chemical composition as the modified slurries. Polishing tests were conducted following the procedure outlined in Chapter 3. To measure the repulsive force barriers created by the self-assembled surfactant structures, a Digital Instruments Nanoscope III Atomic Force Microscope was used. Surfactant solutions were prepared at the same concentration and ionic strength as the polishing slurries and experiments were conducted in a liquid cell. The p-type PECVD silica wafer was used as the substrate. A 1.5 m size silica particle was attached to the cantilever tip to simulate the silica-silica interactions in the CMP applications. The normal force was measured as the substrate was moved towards and away from the particle by the motion of a piezoelectric scanner. The deflection of the cantilever was monitored by a laser that reflects from the top of the cantilever onto a position sensitive photodiode [Duc92]. Maximum repulsive force was determined as the magnitude of the repulsion, just prior to the silica particle jumping into contact with the wafer surface [Adl00]. The same set-up was also used for measuring the friction forces at single particle-substrate interaction level utilizing AFM. The lateral force was measured by
74 engaging a 7 m silica particle (attached to the cantilever tip) to the wafer surface and sliding the substrate perpendicularly to the cantilever. The particle was made to slide over a given location on the silica substrate by disabling the slow scan direction. Figure 5.1 schematically illustrates the method of friction force measurement by AFM technique. The magnitude of the lateral friction force was determined from half the difference in the lateral detector signal in the forward and reverse directions. By systematically increasing the applied load, friction force response was measured as a function of the normal force. The in-situ friction force measurements were conducted by mounting a lateral force measuring instrument to the Struers Rotopol 31 polisher (Figure 5.2). Friction forces were measured at a 3.5 psi down pressure and 150 rpm platen speed. The wafer size was 0.8 x 0.8 inch 2 and an IC-1000 polishing pad was used similar to the one in the polishing experiments. To measure the frictional forces, a Sensotec model load cell was Figure 5.1. Schematic illustration of AFM force measurement technique.
75 Figure 5.2. Schematic illustration of in-situ force measurement technique [Mah99a]. employed. Its output was connected to a data acquisition system and data recorded every 250 ms for a duration of 60 seconds [Mah99a]. To investigate the pad-substrate interactions as a function of applied head pressure, Fourier Transform Infrared Spectroscopy/Attenuated Total Internal Reflection Spectroscopy (FT-IR/ATR) technique was selected. A nitrogen-purged Nicolet Magna 760 spectrometer equipped with a DTGS detector was used to conduct FT-IR analysis. All the spectra were the results of 512 co-added scans at a resolution of 4 cm -1 . The background spectrum for all the experiments was the single-beam spectrum of the dry ZnSe crystal. Figure 5.3 schematically represents the experimental set up. A representative piece of pad sample (IC-1000) was placed on the (ZnSe) ATR crystal and loaded at 3.4, 5.4, 7.4, 9.4 and 11.4 psi pressure. The intensities of the CH 2 peaks of the polyurethane (polishing pad material) were recorded at 3000-2750 cm -1 wave number range by measuring the area under the detected peak. Five spectra were collected at each pressure level. To determine the percent pad contact area at a given pressure, intensity of a100% pad contact was measured. To provide 100 % contact, a defect free pad sample
76 was prepared and its CH 2 peak intensity was measured. To make sure that it was in a complete contact with the ATR crystal, the peak intensity was detected as a function of applied load until the area under the peak remained constant with increased load. To determine the download applied per a single abrasive, concentration of particles on the pad surface contacting the wafer surface was determined. To calculate the number concentration of the particles, Scanning Electron Microscopy (SEM) images of the pad surface were taken after polishing. FTIR-ATR analyses were also conducted to compare the CTAB adsorption on silica as a function of slurry ionic strength. A silicon ATR crystal with silica layer on the surface was used for this analysis, on which 256 co-added scans were collected at a resolution of 4 cm -1 for all the solution conditions. The intensity of the CH 2 peaks of the C 12 TAB surfactant (at 32 mM concentration) was recorded at 3000-2750 cm -1 wavelength ranges. The background spectrum was the single-beam spectrum of the dry silicon crystal with a silica layer on it. The relative adsorption intensity was analyzed by comparing the peak heights of the obtained spectra. Figure 5.3. Schematic representation of the FTIR-ATR technique used to measure the percent pad contact as a function of down load.
77 Results and Discussion Dispersion of CMP Slurries by Controlling Particle-Particle Interactions Stability of the silica suspensions in the presence of 0.1 M NaCl at pH 4 has been investigated previously in our laboratory as a function of the C 12 TAB surfactant concentration [Adl00, Sin01]. Increasing the surfactant concentration from zero to 50 mM, a significant jump was recorded in the repulsive force barrier at a concentration range of 8 to 10 mM. The coagulated silica suspension was also observed to stabilize at this critical concentration range. This correlation between the repulsive force barrier and the slurry stability has been explained based on the formation of self-assembled surfactant structures at the solid-liquid interface [Man 94, Man 95, Adl00]. The separation distance, where the jump occurred was shown to match the size of the micelle confirming the presence of self-assembled aggregates at the solid/liquid interface [Aks96]. To study the stabilization of the silica CMP slurries in the presence of 0.6 M NaCl at pH 10.5, C 12 TAB surfactant was used at 1, 8 and 32 mM concentrations being below, at and above the transition concentration between the unstable and stable suspensions [Adl00, Sin01]. Initially, the baseline polishing slurry was stable due to the presence of high negative charges around the silica particles at pH 10.5. However, CMP operations are conducted in the presence of reactive additives such as oxidizers and complexing agents, which results in slurry destabilization [Ste97]. Therefore, to simulate the severe environments of the CMP slurries, 0.6 M NaCl was added to the silica slurry resulting in coagulation by screening the surface charge. As a result, the mean particle size of the slurry increased from 0.2 m to 4.3 m. When 1mM C 12 TAB was added to the coagulated slurry, it remained unstable with a mean particle size of 4.8 m. However, at
78 Table 5.1. Summary of polishing performance of the baseline (with and without NaCl), C 12 TAB and C 8 TAB mediated slurries in the presence of 0.6 M NaCl. 8 mM C 12 TAB concentration, where a jump was observed in the repulsive force barrier due to the formation of the self-assembled surfactant aggregates, the silica slurry started to stabilize and the mean size decreased to 1.6 m. At 32 mM C 12 TAB concentration, the polishing slurry was completely stable (mean particle size 0.2 m) since an adequate repulsive force barrier was reached under these conditions. Polishing experiments were conducted to evaluate the performance of the slurries stabilized in the presence of C 12 TAB surfactant. Table 5.1 summarizes the stability and polishing performance for the C 12 TAB mediated slurries. Surface quality response was
79 analyzed in terms of surface roughness and maximum depth of the surface defects (pits and scratches) by imaging the polished wafer surfaces using AFM. Slurries containing 0.6 M NaCl resulted in 2.7 nm surface roughness and 70 nm deep pitting as compared to 0.85 nm and 35 nm obtained with the baseline slurry. The poor surface quality in the presence of salt was due to the coagulation of the baseline slurry with the salt addition [Bas02]. Since the 1mM and 8mM C 12 TAB concentrations were not able to stabilize the slurries completely, surface quality was observed to be unacceptable at these concentrations. However, when complete stabilization was reached with the addition of 32 mM C 12 TAB, surface roughness reduced to 0.5 nm and the maximum surface depth of the pits was recorded to be only 20 nm. To evaluate the overall polishing performance, material removal rate response of the slurries was also measured simultaneously with the surface quality response. Addition of 0.6 M NaCl resulted in 7058 /min material removal rate as compared to the 4300 /min with the baseline slurry due to the decreased electrostatic repulsion in system and attack of cations enhancing silica dissolution as discussed in detail in the previous chapter. When 1mM C 12 TAB was added to the high ionic strength slurry, the material removal rate decreased to 5500 /min as compared to baseline result of 7058 /min in the presence of salt. On the other hand, at 8 mM and 32 mM C 12 TAB concentrations, a negligible removal rate of only 88 and 66 /min was obtained, respectively. This observation indicated that in stabilization of the CMP slurries, control of particle-particle interactions was necessary but not sufficient. To design adequately performing slurries, particle-substrate interactions must also be analyzed when the slurry is chemically modified to achieve stabilization.
80 Control of Pad-Particle-Substrate Interactions for Surfactant Mediated Slurries Two reasons can be suggested for the negligible material removal in the presence of 8 and 32 mM C 12 TAB surfactants in the CMP slurries. First, the high repulsive force barrier induced by the C 12 TAB self-aggregated structures may be preventing the particle-surface engagement and hence resulting in a very low material removal rate. Secondly, it is known that the presence of surfactants in the suspensions can result in lubrication between the abrasive and surface to be polished, decreasing the frictional forces [Bhu95b, Xia96]. Therefore, the addition of C 12 TAB in the polishing slurries at relatively high concentrations might have resulted in negligible material removal by reducing the frictional interactions required for polishing. The concentration of C 12 TAB, where negligible material removal rate response was obtained (8 mM) coincided with the observation of the jump in the maximum repulsive force barrier. Therefore, initially, the impact of cohesiveness of the surfactant aggregates on material removal was investigated. Figure 5.4. Schematic representation of the particle-particle and particle-substrate interactions for the silica-silica polishing systems in the presence of self assembled surfactant aggregates.
81 Figure 5.5. Maximum repulsive force response of C12TAB, C10TAB and C8TAB surfactants at 32, 68 and 140 mM concentrations in the presence of 0.6 M NaCl at pH 10.5 obtained with AFM using 1.5 m particle attached to tip. Effect of self assembled surfactant aggregates on particle-substrate interactions Figure 5.4 schematically illustrates the suggested role of the surfactant aggregates in particle-particle and particle-substrate interactions. Cohesiveness of the surface aggregates was observed to result in adequate repulsion between particles providing stability even at high ionic strength conditions. However, it is also possible that these surface aggregates can prevent the particle-substrate engagement during the polishing process, resulting in minimal material removal rates. To determine their impact on particle-substrate interactions, magnitude of the rigidity within the self-assembled
82 surfactant structures needed to be altered. Previous investigations have demonstrated that this can be accomplished by decreasing the surfactant chain length [Adl00]. Accordingly, the force barriers (a measure of the cohesiveness between surfactant molecules constituting the surface structures) of smaller chain length CTAB surfactants were measured by AFM. Figure 5.5 shows the force barriers obtained with baseline (pH=10.5 with and without 0.6 M NaCl) and C 12 TAB, C 10 TAB and C 8 TAB surfactants at 32, 68 and 140 mM concentrations (above CMC concentrations, at pH=10.5 with 0.6 M NaCl). The magnitude of the repulsive force created on a 0.2 m particle by C 12 TAB, C 10 TAB and C 8 TAB surfactants at the given concentrations were 6, 2.2 and 0 nN, respectively. As it was expected, decreasing the chain length of the surfactant led to a smaller repulsive barrier. The C 8 TAB surfactant did not exhibit a repulsive force barrier. Therefore, it was predicted that the slurries prepared with C 8 TAB should result in polishing. To investigate this hypothesis, experiments were conducted with slurries containing 1 mM, 35 mM and 140 mM C 8 TAB (concentrations corresponded to 1, 8 and 32 mM of C 12 TAB). Slurry performance in the presence of C 8 TAB is summarized in Table 5.1. Due to the lack of interparticle repulsion, none of the slurries prepared using this surfactant were stable. Although, the material removal rates were high (5000-7000 /min), the maximum surface deformation values were higher than desired due to presence of coagulates (50-60 nm). With C 10 TAB surfactant addition, stability was reached at 68 mM concentration but the material removal rate remained at 650 /min, which was lower than the baseline value. It was observed that the stability responses obtained as a function of surfactant chain length strongly correlated with the measured repulsive force values (a measure of the strength of the surfactant aggregates). To
83 evaluate the impact of the repulsive force barrier on particle-substrate interactions, it was necessary to know if the surfactant structures were broken during polishing. This required the determination of exact force applied on a single abrasive particle during CMP applications. Force per particle during polishing. To calculate the force applied per particle, it was necessary to determine (i) the wafer pad contact area at the operational pressure of 7 psi and (ii) the particle concentration at the pad surface interacting with the wafer. The IC-1000 polishing pad has a porous surface as demonstrated in Chapter 2 (Figure 2.5). Therefore, only some portion of it will be in contact with the wafer surface during polishing. To determine the percent pad contact area as a function of the applied load, FTIR-ATR technique was used as detailed in the experimental section and illustrated in Figure 5.3. The IR spectra of the pad sample was collected as a function of the applied load. Figure 5.6 a shows the spectra obtained on the pad sample at downloads from 34 to 114 N, corresponding to 3.4 to 11.4 psi pressure. The signal intensity increased with the applied load indicating that more pad contact was achieved with the ATR crystal. To obtain the percent pad coverage, same analyses were also conducted on a defect free piece of polyurethane, which is the material the pad is made of. To achieve 100 % contact, the spectrum was collected on this piece also as a function of pressure. It was observed that above 15 psi pressure, the intensity of the collected spectra remained the same indicating that the 100% contact was reached (Figure 5.6 b). Comparing the absorption values obtained with the porous polishing pad and the defect free pad surface (Figure 5.6 a and b), absorbance values show that the intensity of pad contact was much smaller relative to the.
84 Wavelength (cm -1 ) Abs 3000 2900 2800 0.004 0.003 0.002 0.001 0.000 -0.001 -0.002 11.4 PSI 9.4 PSI 7.4 PSI 5.4 PSI 3.4 PSI (a) Abs 18 PSI 15 PSI 12 PSI 9 PSI 7 PSI 0.70 0.60 0.50 0.40 0.30 0.20 0.10 -0.00 3000 2800 Wavelength (cm 1 ) (b) Figure 5.6. FTIR-ATR spectra for IC-1000 polishing pad. (a) Porous pad contact as a function of applied load; (b) Pore free, 100 % pad contact as a function of applied load.
85 100% pad contact. Indeed, when the percent pad contact was calculated as a function of applied head pressure, the values were calculated to vary between 0.25% to 0.5% in the selected pressure range (Figure 5.7). At the operational pressure of 7 psi, the pad-wafer contact area was determined to be 0.33%. Although this appeared to be a small value, it was in a good agreement with the literature findings, which predicted the pad contact area to vary between 0.05% (hard pad with elastic modulus value E=100 MPa) and 0.54% (soft pad with E=10 MPa) at 7psi [Yu94]. The elastic modulus value of the IC 1000 polishing pad is predicted to be 30-40 MPa [Gun85, Tic99], which is in between the values reported for the hard and the soft pads. In agreement, the experimentally determined percent contact area for IC-1000 pad (0.33%) is also in between the values reported for the hard and the soft pads (0.05-0.54 %). These observations suggested that only a small portion of the pad was in contact with the wafer during polishing. 0.55 0.33 % pad contact 0.50 0.45 0.40 % Pad Contac t 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 8 10 12 0 2 4 6 7 Head Pressure (psi) Figure 5.7. Percent IC 1000 pad contact as a function of applied load.
86 Figure 5.8. SEM image of the IC-1000 polishing pad surface after polishing with 0.2m size 12 wt% polishing slurry, abrasive particles are seen on the pad surface. The second challenge was the determination of the number of particles on the pad surface in contact with the substrate at a given pressure. In order to calculate the number concentration, a representative piece of pad sample was analyzed by Scanning Electron Microscopy (SEM) after polishing with the baseline slurry. Figure 5.8 illustrates the SEM image of the pad surface, which was mostly coated by the 0.2 m particles. To determine the particle number concentration, 25 SEM images were collected and the area coated by the particles was quantified utilizing Image Pro optical analysis software. The average number of particles in contact with the surface was found to be 42 x 10 6 per one inch square area. Dividing the applied download (7 psi = 31.14 N/inch 2 ) to the number of
87 particles, the load applied per particle was calculated to be 750 nN. Since the measured force barriers for all the CTAB surfactants (0, 2.2 and 6 nN for chain lengths 8, 10 and 12, respectively) were much smaller than the calculated load per particle, the self-assembled CTAB aggregates were considered to be broken during the CMP applications. The minimum pressure per particle can be achieved when the pad area is totally coated by the abrasive particles at hexagonal close packed arrangement. Even under this condition the load per particle was calculated to be 500 nN. Furthermore, the impact of pressure on abrasive concentration was analyzed by imaging the pad surfaces polished at 3.4 to 11.4 psi range. The average number of particles did not change significantly with the change in pressure. This was also concluded in the literature by in-situ monitoring of the slurry transport during CMP [Cop00]. Hence, it appeared that, strength of the surfactant aggregates was not the primary reason for the significant decrease in the material removal rates. Surface lubrication effect on particle-substrate interactions To investigate the impact of surface lubrication on material removal rate, frictional forces were measured at system level by utilizing in-situ frictional force measurement and by AFM at the single particle-wafer interaction level. In-situ friction force measurement Figure 5.9 compares the material removal rates obtained with the baseline (with and without NaCl) and surfactant mediated slurries (with 0.6 M NaCl) to the in-situ frictional force values. A direct correlation was observed between the material removal rates and the measured frictional forces. The baseline slurry without NaCl addition resulted in 3.4 N friction force, which increased to 7.2 N in the presence of 0.6 M NaCl. The highly coagulated C 8 TAB modified slurry (at 140 mM) also resulted in high
88 frictional forces (4.5 N), which polished at 5167 /min rate. For the C 10 TAB (68 mM) and C 12 TAB (32 mM) mediated slurries, on the other hand, friction force decreased to 2 N, which was ~4 times lower compared to the baseline value in the presence of 0.6 M NaCl. Although this decrease agreed with the reduction in the material removal rates, it was not as significant as the ~10 times reduction for the C 10 TAB, and 100 times reduction for the C 12 TAB mediated slurries as compared to the polishing with 0.6 M NaCl containing slurry. This inconsistency can be explained based on the fact that this technique does not only represent the particle substrate interactions, but the total frictional force in the polishing system, which is composed of the combined particle-particle, pad-particle, particle-substrate and pad-substrate frictional interactions. Figure 5.9. In-situ friction force and material removal rate responses of the baseline slurries (12 wt%, 0.2 m primary particle size) and the slurries containing C 12 TAB, C 10 TAB and C 8 TAB surfactants at 32, 68 and 140 mM concentrations in the presence of 0.6 M NaCl at pH 10.5.
89 Friction force measurements with AFM In order to understand the surfactant lubrication effect on material removal rate at particle-wafer interaction level, AFM friction force measurements were conducted. A silica particle attached to the cantilever tip was made to raster on the silica wafer surface in the presence of surfactants to simulate the single particle-wafer interaction. Figure 5.10 a shows the lateral versus normal force response of the baseline (pH=10.5) and surfactant mediated solutions without salt addition (140, 68 and 32 mM, C 8 TAB, C 10 TAB and C 12 TAB). It can be seen that for baseline solution, frictional forces increased with the increasing normal forces. On the other hand, in the absence of salt, all the surfactant mediated slurries resulted in minimal friction. Indeed, when polishing experiments were conducted at the given concentrations of these surfactants in the absence of NaCl, the material removal rate values were observed to be negligible for all the chain lengths (61, 53 and 56 /min for C 8 TAB, C 10 TAB and C 12 TAB, respectively). This result indicated that at any chain length, CTAB surfactant was able to form a lubrication layer in the absence of salt. Consequently, minimal particle-surface interactions occurred for these systems resulting in negligible material removal. It is also important to note that all the slurries were stable without salt addition. The polishing results in the presence of 0.6 M NaCl, however, showed an increase in the material removal with decreasing chain length of the surfactant (66, 650 and 6167 /min for C 12 TAB, C 10 TAB and C 8 TAB, respectively). When AFM friction force measurements were conducted in the presence of salt, an interesting behavior was observed. It can be seen in Figure 5.10 b that, C 8 TAB and C 10 TAB mediated solutions started to exhibit higher friction values above750 nN. This type of a behavior was also
90 1500 1200 600 900 Loading Force (nN) 300 0 0 p H 10.5, (Without NaCl) 68 mM C10TAB 32 mM C12TAB Baseline 140 mM C8TAB 200 Lateral Force (nN) 150 100 50 (a) 50 0 Baseline 140 mM C8TAB+0.6M NaCl 68 mM C10TAB+0.6M NaCl 32 mM C12TAB+0.6M NaCl p H 10.5, (With NaCl) 200 Lateral Force (nN) 150 100 0 300 600 900 1200 1500 Loading Force (nN) ( b ) Figure 5.10. AFM friction force measurements on silica wafer with 7m size particle attached to the tip. (a) Solutions containing C12TAB, C10TAB and C8TAB surfactants at 32, 68 and 140mM concentrations without NaCl at pH 10.5; (b) In the presence of 0.6 M salt in the solution
91 observed on the mica surfaces modified with 2, 5, 7 and17 chain lengths alkylsilanes [Xia96]. It appears that the lubricating surfactant layer is desorbed/destroyed beyond a certain loading force. In the absence of salt, all the surfactants, regardless of the chain length, form a compact adhesion layer due to the electrostatic interaction between the negatively charged silica surface and positively charged surfactant head group. However, addition of salt resulted in competitive adsorption of the salt molecules, weakening the adsorbed surfactant and hence possibly desorption of surfactant lubrication layers beyond a certain applied load [Che92]. The impact of competitive adsorption of the salt molecules was less effective on the longer chain length surfactants, perhaps due to their ability to form more densely packed and well-ordered layers [Xia96]. As the length of the hydrocarbon chain increases, the lateral interactions between the hydrocarbon chains become more pronounced resulting in formation of more compact layers. In addition, the driving force leading the surfactant to the substrate surface also increases with the increased chain length, resulting in denser adsorption of the surfactant [Cle99]. Thus, it is possible that NaCl addition does not affect the lubrication layers created by C 12 TAB mediated slurries to the same extend as surfactants with shorter chain lengths as seen in Figure 5.10 b. Accordingly, C 12 TAB yielded negligible material removal of 66 /min, whereas the shorter chain length C 10 TAB surfactant, resulted in material removal of 650 /min indicating that the silica particles were able to engage with the silica wafer surface up to some extend due to the removal of the surfactant. Finally, C 8 TAB mediated slurries also showed an increase in the frictional forces at single particle-surface interaction level, suggesting that they as well should polish the silica surface. However, the significantly high removal rate for the C 8 TAB mediated slurries (5167 /min) should be attributed not
92 only to easier removal of the loosely packed 8-carbon chain surfactant layer but also to the coagulation of the particulates in the absence of a repulsive force barrier for C 8 TAB at 0.6 M salt concentration. Modification of the Interaction Forces by Calcium Addition It is indicated in the previous section that ability of the adsorbed surfactant to form a lubricating layer is adversely affected by Na + addition to the system. The Na + ions tend to approach the negatively charged silica surface at high pH values due to electrostatic interactions [Bud61]. Furthermore, it has been shown that in the case of silica, stability constants characterizing the intensity of interaction between surface hydroxyl and the cation increases with the increasing valency of the added salt [Sjo96]. This observation suggested that, by varying the valency of the added salt, the frictional forces could be manipulated. To investigate this hypothesis, 0.24 M CaCl 2 was added into the polishing slurries (the ionic strength was kept equivalent to 0.6 M NaCl). Table 5.2 summarizes the polishing performance in the presence of CaCl 2 . It was observed that the slurry was stable and hence the surface quality of the polished wafers was acceptable with roughness value of 0.47 nm, and the maximum depth of the observed defects less than 16 nm. Moreover, the material removal rate was high even in the presence of 32 mM C 12 TAB surfactant (4800 /min). The speciation diagram of calcium at pH 10.5 shows that Ca ++ is the predominant specie in the solution with CaOH + [Fue95]. The divalent Ca ++ ions are more competitive in adsorbing on the silica surface as their intensity of interaction on the negatively charged silica surface is higher than Na + ions [Sjo96]. Specifically, it was shown that adsorption of calcium species on quartz extensively increase starting at pH 10 [Cla68]. It was also suggested that the adsorption of calcium hydroxyl complex on the
93 Table 5.2. Summary of polishing performance of the baseline (with and Without CaCl 2 ), and C 12 TAB mediated slurries in the presence of 0.24 M CaCl 2 . quartz surface create a positively charged active site, leading to increasing recovery in mineral flotation process [Ye93]. Therefore, the adsorption of the cationic surfactant is expected to be much less on the calcium activated silica surface resulting in weaker lubrication. Figure 5.11 illustrates the FTIR absorbance spectra for the solutions containing 32 mM C 12 TAB at pH 10.5 without any salt addition and in the presence of 0.6 M NaCl and 0.24 M CaCl 2 . It can be seen that the most intense CH 2 peak was obtained in the absence of salt, indicating effective surfactant adsorption. The frictional forces were recorded to be negligible for this system (Figure 5.10 a). Addition of NaCl reduced the intensity of the peak, which was attributed to the competitive adsorption of the Na + ions. CaCl 2 addition resulted in the most significant decrease, confirming that calcium species reduced the C 12 TAB adsorption significantly. In agreement, the direct measurement of the C 12 TAB (32 mM) adsorption on the silica surface in the absence of salt gave 6.5x10 -6 mol/m 2 value, while adsorption decreased to 5.2x10 -6 and 1.4x10 -6 mol/m 2 with 0.6 M NaCl and 0.24 M NaCl 2 addition, respectively.
94 32 mM C12TAB 3000 0.012 Abs 0.001 0.000 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.010 0.011 2950 2900 2850 32 mM C12TAB + 0.24 M CaCl2 32 mM C12TAB + 0.6 M NaCl Figure 5.11. FTIR-ATR adsorption spectra of the solutions in the presence of 32 mM C 12 TAB without salt and in the presence of 0.6 M NaCl and 0.24 M CaCl 2 . Figure 5.12 a shows the repulsive force barriers obtained with the baseline and 32 mM C 12 TAB mediated slurries in the presence of NaCl and CaCl 2 . It can be seen that in all the cases there was a repulsive interaction resulting in the stability of the slurries. In case of CaCl 2 addition, repulsive forces were observed at much smaller separation distances as compared to the NaCl added slurries. The repulsive force barrier observed with the CaCl 2 added slurries was also relatively smaller than in the presence of sodium. This may again be attributed to the reduced surfactant adsorption resulting in smaller number of surfactant aggregate existence on the silica surfaces for calcium mediated systems. Slurry stabilization at smaller separation distances is another indication of the increased surfactant removal from the surface, leading to particle-substrate interaction.
95 40 p H 10.5 32 mM C12TAB+0.24M CaCl2 32 mM C12TAB+0.6M NaCl Baseline Force/R (mN/m) 30 20 10 0 -10 0 5 Separation (nm) 10 15 20 (a) Baseline p H 10.5 32 mM C12TAB+0.24M CaCl2 32 mM C12TAB+0.6M NaCl 400 Lateral Force (nN) 300 200 100 0 0 300 Loading Force (nN) 900 600 1200 1500 (b) Figure 5.12. AFM force measurements for C 12 TAB mediated slurry in the presence of NaCl and CaCl 2 . (a) Repulsive force measurements; (b) Friction force measurements.
96 The friction force measurements conducted in the presence of 0.24 M CaCl 2 for C 12 TAB mediated slurries also illustrated high frictional interactions, which was consistent with the high material removal rates obtained on this system. The in-situ friction force was measured to be of 6.3 nN. Furthermore, the single particle-substrate level interactions also showed high friction response as illustrated in Figure 5.12 b. In lubrication literature, the surfactant layers are designed to be strong enough to prevent the friction of the metal or ceramic pieces of the machinery during operation. The load required for destroying the cohesive lubrication layer and resulting in contact of the bare surfaces is defined as the scuff load [Ask86, Kum90]. In CMP however, it was required to modify the strength of the lubricating layer to achieve sufficient frictional interactions and material removal. Based on these findings, it is suggested that consistently high performing slurries could be achieved by controlling the interaction forces in CMP. It was also demonstrated that, control of the interaction forces is possible by manipulating the slurry chemistry utilizing surfactants and varying the ionic strength. Summary Robust dispersion of the CMP slurries in extreme ionic strength and high pH environments is a must for optimal polishing performance, which requires the introduction of high enough repulsive forces between the slurry particulates. However, in this study, it was observed that, to design optimally performing CMP slurries, the control of particle-particle interactions is not sufficient as it maybe in other applications. To enable an optimal material removal rate, it is also necessary to have the abrasive particles engage with the substrate. It has been shown that the particle-substrate interaction forces could be modified by addition of surfactants to the slurries, which tend to form a lubrication layer on the wafer surface and vary the material removal response. The
97 adsorption/desorption of the surfactant molecules were observed to be a function of the slurry ionic strength and the applied normal forces. In summary, it was observed that the stability of the CMP slurries must be tailored by controlling both the lateral (particle-particle) and normal (particle-wafer surface) interactions, which can be manipulated by varying the surfactant chain length, slurry ionic strength and the valency of the added salt. The particle-substrate interaction in CMP is also a function of the slurry particulate properties. In the following chapter, effects of the slurry particle size and solids concentration on CMP performance are investigated.
CHAPTER 6 DESIGN CRITERIA FOR OPTIMALLY PERFORMING CMP SLURRIES Introduction In previous chapters, effects of slurry particle size distribution and stability on CMP performance were investigated. Role of particle-particle and pad-particle-substrate interactions in developing optimally performing CMP slurries was delineated. The next step in formulating engineered particulate systems for effective polishing is the determination of slurry particulate properties, such as the abrasive particle size and solids concentration. The earlier studies on slurry particulate properties reported inconsistent results stating that, the material removal rate increases, decreases or does not change with the slurry particle size [Bro81, Jai94, Izu79, Coo90, Siv92, Bie98]. This is believed to be due to the variation in the selected operational regimes and polishing systems. Indeed, a systematic study on silica CMP has shown that all three of these behaviors could be observed in silica polishing depending on the selected slurry particle size and solids concentration range [Mah00]. Thus, it is clearly required to investigate the mechanisms of polishing as a function of the slurry particulate properties, to design optimally performing slurries. In this chapter, significance of the changes in the slurry particulate properties on polishing performance is investigated for silica CMP process. The mean particle size and solids loading of the abrasives in the polishing slurry may impact the material removal rate and quality of surface finish, as these factors alter the load applied per abrasive 98
99 particle. Being another factor that influences the pad-particle-substrate interactions, the influence of applied download on polishing performance was also evaluated. The obtained results were analyzed in terms of the pad-particle-substrate interactions, which highlighted the possible mechanisms of polishing in silica CMP. Furthermore, a response surface design was conducted to optimize the polishing performance based on the slurry particulate properties and the operational pressure value according to the desired performance criteria. Experimental In this study, the material removal rate and surface quality responses of the slurries of different particle sizes were analyzed as a function of solids loading and applied download. Furthermore, a rotatable central composite design was utilized to optimize the slurry particulate properties at various operational head loads. The central composite design has the advantage of analyzing three factors at five levels with a relatively small number of observations, that are enough to estimate the second order effects of the response surface [Mas89]. It is also suitable for process optimization. Figure 6.1 shows the three-factor layout for the rotatable central composite design. On this layout, the high and low levels of the design are coded as -1 and +1 and the factorial portion is represented with the circles, which are forming a box. The star () points are the axial points, and they are located 1.68 coded units away from the center ( = [2k] 1/4 , for k=3 = 1.68), making the design rotatable. A rotatable design provides equally good predictions at points of equal distance away from the center. Design Expert software (by Stat-Ease Inc.) was utilized to create the response surface design and to analyze the experimental results. The design required 20 experimental runs including 6 repeat tests at
100 Figure 6.1. Central composite design for three factors. the center point, which provided an estimate of the uncontrolled experimental error variation. The studied design and the obtained responses are given in the appendix with the mathematical equations showing the dependence of the selected factors to the experimental variables. Polishing slurries were prepared using monosized sol-gel silica powders obtained from Geltech Corporation by following the procedure described in Chapter 4. Table 1.1 summarizes the selected design levels determined based on the industrially used values. Polishing tests were performed on 1.5 inch x 1.5 inch p-type Table 6.1. Selected design levels for the three-factor rotatable central composite design. Design Levels -1.68 -1 0 1 1.68 Particle Size (m) 0.2 0.3 0.5 0.8 1.0 Applied Load (N) (psi) 34 3.4 54 5.4 74 7.4 94 9.4 114 11.4 Solids concentration (%wt) 0.5 6 15 24 30
101 Figure 6.2. Prediction profiles for material removal rate response. silicon wafers on which a 2.0 m thick SiO 2 layer had been deposited by PECVD. Struers Rotopol 31 polisher and IC 1000/Suba IV stacked pads were used for polishing as outlined in Chapter 3. Results and Discussions Material Removal Rate Response Figure 6.2 shows the prediction profiles obtained from the central composite design analysis for the material removal rate response, as a function of the slurry particle size, solids concentration and applied head load. The R 2 value was obtained from the statistical analysis for the material removal rate response was 0.95 indicating a sufficient experimental fit. Increasing the particle size and solids concentration of the slurry resulted in a decrease in the material removal rate. On the other hand, increasing the applied load led to a tendency for increased material removal. Statistical analysis showed that the slurry particle size and solids concentration had a significant effect on the material removal rate response within 90 % confidence interval. The applied load, however, had only a modest affect on the material removal rate at the selected load
102 Material removal rate Applied load (N) Solids Concentration (%wt) 15.00 22.50 30.00 0.00 7.50 74.00 94.00 114.00 54.00 34.00 3008.99 2420.46 1831.94 1243.42 654.89 Figure 6.3. Material removal rate surface response for 0.5 m particle size slurry. regime. The three-dimensional surface response plot of 0.5 m particle size slurry is illustrated in Figure 6.3. The shape of the contours verify that at a given solids concentration, the increase in the applied load does not have a major impact on material removal rate. To further understand the reasons for the observed dependence of material removal rate to the experimental variables, additional experiments were conducted by varying one factor at a time, as discussed below. Effect of slurry particle size and concentration on material removal rate Figure 6.4 shows the material removal rate responses of 0.3, 0.5 and 0.8 m size abrasive particles as a function of the slurry solids loading at 7.4 psi head pressure. It was observed that for all the selected particle sizes, there was a particular solids concentration, where the material removal rate reached a maximum, beyond which, removal rate started to decrease and reached a plateau around 24 wt% concentration.
103 Mahajan also observed a similar behavior for the silica-silica polishing system previously for 0.2, 0.5, 1.0 and 1.5 m particle size slurries at a solids concentration range of 0.2 to 15 wt%, except for the 0.2 m size slurry, which showed a continuous increase in the material removal rate up to 10 wt% and then reached a plateau (Figure 2.14) [Mah00]. The initial increase in the material removal rate as a function of solids concentration was not observed in the central composite design response, since there was only one data point below 6% concentration region. This data point, being out of the expected response range, was detected as an outlier in the statistical analysis, which was taken out to obtain a reliable experimental fit. To investigate the reasons for the variation in the material removal rate as a function of solids concentration, surface quality analyses were conducted on the wafers Figure 6.4. Material removal rate analysis for 0.3, 0.5 and 0.8 m size slurries as a function of solids loading.
104 polished with 0.3 m size slurries at 7.4 psi. The AFM surface micrographs obtained on these surfaces are presented in Figure 6.5 with the schematics of the suggested particlesubstrate interaction mechanics. At 0.5 wt% solids concentration, many scratches were detected on the wafer surface. At this concentration, only a few abrasive particles were in interaction between the pad and the wafer as confirmed by the SEM micrographs taken on the pad surface after polishing (Figure 6.6). The number of particles detected on the pad surface also decreased as the particle size increased (in accordance with the lower number concentration of larger particles at a constant solids concentration). Hence, at 0.5 wt%, the normal load applied per particle could be relatively high, which may result in deeper indentation of the particles into the wafer surface as suggested by the indentation mechanism discussed in Chapter 2 (Figure 2.15 b). In addition, abrasive particles are expected to indent into the polymeric polishing pad as schematically illustrated in Figure 6.7 [Qin03]. When the particles are embedded into the polymeric pad, it is anticipated that they may be sliding on the wafer. The scratches observed on the AFM image of the wafer polished with the 0.5 wt% slurry indeed supported the suggestion that the particles were sliding on the wafer surface during polishing and abrading the wafer surface based on the indentation mechanism. Although this resulted in polishing, the material removal rate was only 235 /min. The relatively low removal was attributed number of particles (which are the cutting tools) in contact with the wafer surface being low at this solids concentration (Figure 6.6). Increasing the particle concentration to 6 wt%, an increase in material removal rate was observed (1700 /min), due to the increased number of polishing particles. AFM analysis conducted on the wafer surface polished with 6 wt% slurry still showed
105 many scratches, suggesting that the particles continued to slide on the wafer surface. At 15 wt%, however, the surface image showed more pitting type of deformations than scratches. Similar pitting type deformations were also observed at 24 and 30 wt% concentrations and the material removal rate response started to decrease above 15 wt% concentration simultaneously with the observed changes in the surface quality of wafers. Figure 6.5. Suggested particle-substrate interactions for 0.3 m size slurries as a function of solids loading with the AFM surface micrographs.
106 (a) 0.3 m (b) 0.5 m (c) 0.8 m Figure 6.6. SEM images conducted on the pad surface after polishing with 0.5 %wt slurries. (a) 0.3 m; (b) 0.5 m; and (c) 0.8 m particle size.
107 Figure 6.7. Pad-particle-substrate interactions as a function of increasing slurry solids concentration [Qin03]. It was initially hypothesized that the decreased force per particle at more concentrated slurries might have decreased the frictional interactions, which are responsible for the material removal in CMP [Coo90, Mah99a, Mah99b]. In agreement, SEM images of the IC-1000 pad surface collected after polishing with 0.3, 0.5 and 0.8 m particles at 30 wt% concentration showed that the pad surface was almost completely covered with the particles for all slurries (Figures 6.8 a to c). Furthermore, based on the surface defects being more pitting type than scratching, it was suggested that the particles might have started rolling on the wafer surface during polishing, which may in turn explain the decreased frictional interactions. To investigate these hypotheses, in-situ friction force measurements were conducted and the possibility of particle rolling was evaluated based on torque calculations on the abrasive particles as a function of the load applied per particle during polishing. Figure 6.9 illustrates the in-situ friction force and material removal rate responses of the 0.3 m slurry as a function of the slurry solids concentration. It can be seen that the decrease in material removal rate response occurred at the same concentration range with the decrease in the in-situ frictional force. A direct correlation between the material removal rate and the frictional interactions was observed previously for silica polishing
108 (a) 0.3 m (b) 0.5 m (c) 0.8 m Figure 6.8. SEM images conducted on the pad surface after polishing with 30 %wt slurries. (a) 0.3 m; (b) 0.5 m; and (c) 0.8 m particle size.
109 0 5 10 15 20 25 30 0 500 1000 1500 2000 20 Material Removal Rate (/min) In-Situ Friction Force (N) 15 10 5 0 Solids Concentration (wt.%) Figure 6.9. In situ friction force measurements of 0.3 m size slurries with material removal rate response as a function of solids concentration. The dotted line illustrates the solids concentration where rolling motion starts according to the developed model. [Mah99a, Mah99b]. Our analysis on the surfactant mediated systems also indicated a correlation between the frictional interactions and the material removal rate as summarized in the previous chapter. To understand the reason for the decrease in the frictional interactions, the second hypothesis, possibility of particle rolling during polishing, was evaluated. To investigate conditions for sliding or rolling of abrasive particles during polishing, a model developed in our group was utilized [Qin03]. This model analyzes the pad-particle-substrate interactions as a function of the slurry particle size and solids concentration, being the factors that impact the load per particle. In this model, the pad-particle and particle-substrate lateral (frictional) interactions creating resistance to particle rolling are compared to the torque on the particle that is generated by the relative movement of the polishing head (carrying the wafer) and the platen (to
110 which the polishing pad is attached). Based on the Hertzian interactions, the particles were determined to start rolling when a critical angle of 36 o was satisfied between the center of the abrasive and the pad surface (Figure 6.10). Accordingly, it was calculated that all the 0.3 m particles will experience rolling motion at 22.4 wt% solids concentration threshold for rolling. This value was calculated to be18.6 wt% and 11.7 wt% for the 0.5 and 0.8 m size slurries, respectively. The torque on the spherical abrasive particle increases with the increasing particle size, since it is directly related to the particle radius. Therefore, the complete rolling motion of the larger particle size systems is expected to be reached at lower solids concentrations as compared to the smaller size particles. The calculations conducted based on the developed model also followed this trend. However, it is important to note that the proposed model predicts the solids concentration at which all the particles are in rolling motion, rather than the Figure 6.10. Schematics of the pad-particle-substrate interactions with the forces induces during polishing. Particle starts rolling when exceeds 36 o [Qin03].
111 concentration where the transformation from sliding to rolling motion starts. Based on the observed changes in the type of surface defects and the material removal rate response, it is believed that this transformation occurs at the concentration, where the material removal rate reaches a maximum and starts decreasing. In the proposed model calculations, the average load per particle is taken into account, which allows for the prediction of the solids concentration where all the abrasives reach rolling motion. In real polishing applications, on the other hand, there is a distribution of the normal force applied per particle. One source of this distribution is believed to be the height distribution of the pad asperities, which transfer the applied head load to the abrasive particles [Yu89]. The AFM cross section of the pad surface represented in Figure 2.5-b illustrates the asperity height difference. In addition, the local differences in the particle concentration on the pad surface could result in variation in the load per particle. Therefore, it is suggested that the particle rolling starts when a decrease in material removal rate is detected. Before this point, the majority of the particles are believed to be in sliding motion on the wafer surface as they are indented on the pad material. After this concentration, however, some portion of the particles may be rolling while the remaining continue the sliding motion until the complete rolling is reached. According to the modeling calculations, at 24 %wt solids concentration, all the selected size slurries are expected to be in complete rolling. In agreement, the experimental measurements showed a plateau in material removal rate above this concentration for all the 0.3, 0.5 and 0.8 m size slurries (Figure 6.4). Furthermore, the material removal rate responses were almost the same regardless of the slurry particle size. This observation was attributed to the saturation of the pad surface with the particles at 24 to 30 wt% solids concentration
112 range as illustrated in Figure 6.9 a to c. According to the Hertzian contact theory, the total area of contact should remain the same for any particle size at constant head pressure, when the surface saturation is reached. The slight tendency to increase in material removal rate at 30 wt% can be explained based on the formation of multiple layers of abrasive particles as illustrates in Figures 6.9 a to c. However, if the complete rolling was not achieved, material removal rate did not decrease as observed for 0.2 m size slurries previously (Figure 2.14), for which the modeling calculations suggested that the complete rolling motion could be reached at 35.5 wt% solids concentration. These analyses indicated that the maximum removal rates were obtained for any particle size slurry, at the suggested point of transition from sliding motion to rolling motion. However, surface quality of the polished wafers must also be taken into account to determine the optimal operational regime. It is believed that in sliding motion, where the load per particle is relatively high, material removal takes place more mechanically, which was defined as the indentation based mechanism previously as discussed in Chapter 2 [Bie98]. This type of material removal also results in higher frictional forces due to the mechanical abrasion of material [Coo90]. As the solids concentration is increased, the number of polishing particles increases leading to higher removal rates till a maximum is reached. Up to this maximum, particles are still indenting and sliding on the surface resulting in mechanical material abrasion leading to highest total indent volume during polishing. Immediately above this concentration rolling motion is suggested to start in which the particle indent is minimal. Therefore, more chemically driven particle-substrate interactions occur, which was defined as the contact area based material removal mechanism in previous investigations [Bie98, Mah00]. This type of
113 interactions was suggested to improve the formation of the chemically modified surface film on the silica wafer [Bie98, Tom94] by increasing the water diffusion into the silica structure [Ito81, Mas84, Dun91]. Although the material removal rate decreased due to the reduced mechanical interactions, the surface quality started improving with the increasing solids concentration since the particle indent was minimal. To balance a high material removal rate and acceptable surface finish during polishing, chemical and mechanical interactions must be tailored by selecting appropriate slurry particle size and concentration. Use of small particle in CMP is therefore advantageous since the small size abrasives tend to start rolling at relatively higher concentrations, where a high enough total indent volume can be obtained leading to high material removal rates with minimal surface defects. In agreement, the commercial CMP slurries are commonly made of 50 to150 nm size particles at 10 to 30 wt% solids concentrations. Effect of applied head pressure on material removal Figure 6.11 illustrates the material removal rate responses of an abrasive free solution, and the 0.3, 0.5 and 0.8 m size slurries at 15 %wt, as a function of applied load. It was observed that without the particles, material removal rate was negligible at any applied pressure indicating the need for abrasives to achieve polishing in silica-silica system in agreement with previous observations in the literature [Izu79]. In the presence of particles, there was a trend towards increased material removal with the increased head load. Since the head load is transferred to the wafer surface by the polishing pad, it was necessary to examine the pad-substrate interactions as a function of the applied load. In Chapter 5, FTIR-ATR technique was used for determining the percent pad contact area with the wafer surface as a function of the operational head load. It was
114 Figure 6.11. Effect of head pressure on material removal rate for abrasive free and 0.3, 0.5 and 0.8 m size slurries. observed that the pad-substrate contact area increased with an increase in the applied load. During polishing, an increase in head load is expected to bring more pad material in contact with the wafer. Therefore, although the applied head pressure is increased, the pressure per particle would not change to the same extend since the number of particles in contact also increases. Accordingly, as verified by the statistical analysis (at 90% confidence interval), the material removal rate response did not vary considerably as a function of applied load, which was also reported in the literature [Cop00]. Surface Quality Response Figures 6.12 and 6.13 show the prediction profiles for the root mean square (RMS) surface roughness and maximum depth of surface defects (pits or scratches)
115 Figure 6.12. Prediction profiles for surface roughness response. Figure 6.13. Prediction profiles for maximum depth of the surface defects (pits or scratches) response. responses as a function of the slurry particle size, solids concentration and applied head load. The statistical analysis indicated that the particle size and the solids loading significantly affected the surface roughness response at a 90% confidence interval (R 2 = 0.94). For the maximum depth of the surface defects analysis, on the other hand, particle size and applied load variables were found to be the statistically significant factors (R 2 = 0.97 R 2 ).
116 It is observed from Figures 6.12 and 6.13 that, increasing the abrasive size resulted in an increase in both the surface roughness and the maximum surface deformation responses. Figure 6.14 a and b show the AFM images and cross section profiles taken on the wafers polished with 0.2 and 1.0 m size slurries at 15 wt% solids concentration under 74 N applied load. These images clearly illustrate the increasing magnitude of surface pitting with increasing particle size. For the small size particles some pittings occurred on the wafer surface but they were relatively shallow. Therefore, the surface roughness was within acceptable range. Increasing the particle size at the same concentration up to 0.5 m size increased the depths of the surface deformations, which also increased the surface roughness response. When the particle size was increased to 1.0 m, there were significantly fewer particles interacting with the wafer surface and, although the depths of the pits increased, their number decreased thus resulting in a decrease in surface roughness. Figure 6.14. AFM pictures and surface profiles for the wafers polished with 15 wt% slurries under 74N applied downforce. (a) 0.2 m size slurry; (b) 1.0 m size slurry.
117 The impact of solids concentration on surface quality was presented earlier in Figure 6.5. At low solids concentrations (0.5 and 6 wt%), the polished surfaces had more scratch type of deformation since the particles were suggested to be in sliding motion. The surface roughness values for 0.3 m size slurries were 0.84 and 1.31 nm, at 0.5 wt% and 6 wt% solids concentrations. Under these conditions, the maximum defect depth did not exceed 27 nm. At 15 wt% solids concentration, however, the surface roughness and the pitting depth were the highest with 1.85 and 42 nm values respectively. This can be attributed to the change in the particle-substrate interactions from sliding to rolling, as explained in the previous section. Particle rolling motion could be the reason for the deep pit formations at this concentration, where the pressure per particle was still relatively high. At increased solids concentrations, surface quality was observed to improve, which may be attributed to the increased number of abrasive particles in contact with the surface thus decreasing the load per particle. Surface roughness was detected to be 1.02 and 0.84 nm, for the 24 and 30 wt% slurries, respectively. The maximum defect depth was measured to be below 30 nm in each case. Finally, the effect of applied load on surface quality was investigated. Increasing the applied load adversely impacted the surface quality. Both surface roughness and deformation responses increased as the head load increased as can be seen on the prediction profiles given in the Figures 6.12 and 6.13. Optimization The overall goal of the CMP process is to achieve a high material removal rate while protecting the wafer surface quality. After the impact of slurry particulate properties and head pressure were investigated on the polishing performance, optimization analyses were conducted in the final part of this study. A central composite
118 design was utilized for optimization purposes. To determine the optimal operational regimes, appropriate limits were defined for each response surface. Material removal rate response was considered to be acceptable above 2000 /min. Surface roughness and maximum deformation responses were restricted to be less than 1.5 nm and 40 nm to have a reasonable surface quality after polishing. Under these conditions the contour plots of the three response surfaces were plotted. Figure 6.15 shows the solids loading versus particle size contour plots at the selected limiting conditions for different applied loads. The colored areas indicate the regions that are restricted by one of the response surface limitation. The white areas, on the other hand, represent the optimal polishing regions where all the desired conditions are satisfied. Figure 6.15 a shows the surface response at low pressure regime, where 54 N (5.4 psi) head load was applied during polishing. In this case, the limiting factors were the material removal rate response and the maximum depth of surface pitting or scratches. As discussed previously, material removal rate tends to increase with increasing applied load, whereas the surface quality is better in low load regimes. Therefore, the desired 2000 /min material removal rate condition could be satisfies at a more limited region than the surface quality requirements. As the applied load increased to 94 N, (9.4 psi), the material removal rate response improved. However, it can be seen in Figure 6.15 b that the surface quality responses started to become the limiting factors as the surface roughness increased significantly. Consequently, the operational area was limited to a small region. The optimal download, where the operational area is maximum, was determined to be 76.2 N according to the Design Expert optimization results as illustrated in Figure 6.15 c.
119 Rmax: 40 RMS: 1.5 RMS: 1.5 MRR: 2000 0.20 0.00 30.00 Solids Concentration (%wt) 22.50 15.00 7.50 0.40 0.60 Particle Size ( m ) 0.80 1.00 (a) 54.00 N Applied load 30.00 Rmax: 40 RM S: 1 .5 MRR: 2000 RMS: 1.5 RMS: 1.5 Solids Concentration (%wt) 22.50 15.00 7.50 0.00 0.20 0.40 0.60 0.80 1.00 Particle Size ( m) (b) 94.00 N applied load Figure 6.15. Continued.
120 RMS: 1.5 Rmax: 40 RMS: 1.5 MRR: 2000 0.20 0.00 30.00 Solids Concentration (%wt) 22.50 15.00 7.50 0.40 0.60 0.80 1.00 Particle Size ( m ) (b) 76.20 N applied load Material removal rate (MRR) is less than 2000 /min. Surface roughness (RMS) is higher than 1.5 nm. Maximum surface deformation is higher than 40 nm. Figure 6.15. Optimal operational regimes as a function of solids concentration and particle size under different applied load conditions. (a) 54 N applied load; (b) 94 N applied load; (c) 76.2 N applied load. According to the optimization analysis, the material removal rate can be increased by increasing the applied load during CMP applications at the expense of surface quality. It is observed that the optimal polishing is achieved at small particle sizes and medium solids loading conditions. These are also the conditions employed in the industrial CMP processes.
121 The statistical analyses also give the advantage of predicting performance of a polishing system within the selected design limits. Equations 6.1 to 6.3, give the quadratic relationships for the estimation of material removal rate (MRR), surface roughness (RMS), and maximum surface deformation (R max ) as a function of the slurry particle size (A), solids concentration (B) and applied head load (C). MRR = 4140 â€“ 6796 Aâ€“ 189 B + 11 C + 4286 A 2 + 5 B 2 + 0.06 C 2 + 89 A B â€“ 16 A C â€“ 0.45 B C (6.1) RMS = 4.38 + 15.11 A + 0.23 B â€“ 5.21 C â€“ 9.07 A 2 â€“ 4.97 B 2 + 2.74 C 2 â€“ 0.09 A B â€“ 0.02 A C â€“ 2.98 B C (6.2) R max = 75.1 + 159.47 A + 0.15 B â€“ 3.1 C â€“ 42.7 A 2 + 0.03 B 2 + 0.02 C 2 â€“ 8.07 A B + 1.79 A C + 0.03 B C (6.3) Summary In this chapter, effects of slurry particulate properties on the performance of oxide CMP operations were investigated by examining material removal rate and surface quality responses as a function of the slurry particle size and solids loading at different applied load regimes. The obtained results were explained by studying the pad-particle-wafer surface interactions based on the material removal rate response and the type of surface deformation obtained on the polished surfaces. Optimization results showed that most desirable polishing results are achieved with small particle size slurries at medium solids concentrations with medium applied loads. These findings are in agreement with the industrial practice. Figure 6.16 summarizes the criteria that must be meet for the formulation of effective CMP slurries. It can be concluded based on the findings of this investigation that, the design criteria for optimally performing CMP slurries are (i) the oversize particles must be avoided in the
122 Figure 6.16. CMP slurry design criteria. slurries, (ii) robust stabilization schemes must be used in CMP by controlling the particle-particle and particle-substrate interactions and (iii) monosized slurries with small particle size at medium solids concentrations should be used for polishing at medium operational
123 load/pressure regimes. These criteria can be used to develop effective predictive methodologies for the CMP applications, which is discussed in the following chapter as a part of the suggested future work.
CHAPTER 7 SUMMARY AND SUGGESTIONS FOR FUTURE WORK Summary The advances in the microelectronics industry continue to demand significant improvements in the chemical mechanical polishing (CMP) processes. As the sizes of the microelectronic devices keep on decreasing, an atomic level material removal with minimal number of defects is targeted in planarization efforts. To meet the desired performance criteria, it is necessary to understand the fundamentals of the pad-particle-substrate interactions leading to material removal and defect formation in CMP rather than concentrating on the tool related operational variables alone. In this study, design criteria were developed for consistently high performing CMP slurries by investigating the particle-particle and pad-particle-substrate interactions during polishing. One of the major concerns in CMP is the formation of scratches and pitting on the wafer surfaces, which may result in defective microprocessors. Although it has been commonly suggested in the literature that the presence of larger particles in the slurries resulted in defect formation, their impact on polishing performance was not clearly known. In the beginning of this investigation, effects of hard-core coarser particles (hard agglomerates) on polishing performance were quantified. The limits of light scattering technique in detecting a small number of coarser particles in a commercial CMP slurry were established. It was observed that the limits of detection could be as low as parts per million (ppm) level. However, the polishing tests conducted in the presence of hard agglomerates at these detection limits revealed that, even at ppm concentrations, they 124
125 resulted in unacceptable surface defects such as scratches and pitting. Furthermore, the material removal rate response was found to vary indicating that the agglomerates resulted in changes in the particle-substrate interactions. Being larger in size, the coarse particles seemed to carry most of the load transferred by the polishing pad, reducing the interaction of smaller size, primary slurry particles with the wafer surface during polishing. This also resulted in unequal distribution of the forces leading to surface deformations as the coarser particles indented to the wafer surface. Therefore, coarser particles must be effectively removed from the CMP slurries to maximize highly uniform particle-substrate interactions leading to good control of material removal and acceptable surface quality. Although filtration of the CMP slurries is commonly practiced to remove hard agglomerates, surface scratches are observed even with the filtered slurries. This has been attributed to the formation of transient (soft) agglomerates of the primary size slurry abrasives. These soft agglomerates may form during polishing due to the fluctuations in the slurry chemistry and particle concentration. To evaluate their impact on polishing performance, partially dry aggregated, polymer flocculated and salt coagulated slurries were prepared. The systematic analysis conducted in the presence of these transient agglomerates confirmed that in spite of their breakage during polishing, soft agglomerates could degrade the surface quality. The magnitude of surface defects depended on the rigidity of the agglomerates. While the relatively strong dry aggregates resulted in severe pitting, loosely bonded polymer flocs created lesser defects. The material removal rate response, on the other hand, was observed to be governed by the extent of the pad-particle-surface interactions that impact the magnitude of the frictional
126 forces enabling abrasion and material removal. Presence of polymer in the slurry decreased the frictional forces by leading to surface lubrication, while the salt mediated slurries were shown to increase the frictional interactions by reducing the repulsive forces between the abrasives and the wafer surface. Eventually, it became clear that not only the hard agglomerates but also the soft agglomerates of the primary abrasive particles must be avoided in the polishing slurries to ensure acceptable polishing performance. Thus, it was required to provide adequate stabilization schemes for the CMP slurries, which could be effective in extreme environments of the polishing process. Robust dispersion of the CMP slurries at high ionic strength and pH conditions in the presence of reactive additives requires repulsive forces between the slurry particulates. In this study, cohesiveness of the self-assembled surfactant structures at the solid/liquid interface was utilized to stabilize CMP slurries. These structures created strong steric repulsive barriers among silica abrasives enabling dispersion at high ionic strengths. However, it was observed that, to design optimally performing CMP slurries, control of particle-particle interactions was a necessary but not a sufficient condition. To provide an optimal material removal rate environment, it was determined that the abrasive particle must engage with the substrate surface. The particle-substrate interactions were observed to be strongly influenced by the added surfactants. Surfactant molecules were determined to form a lubrication layer on the wafer surface altering the frictional forces, which govern the material removal rate. The adsorption/desorption of the surfactant at a given applied load was observed to be a function of the surfactant chain length, slurry ionic strength and valency of the added salt, which led to the control of the particle-substrate interactions and hence the material removal rate.
127 The particle-substrate interactions in CMP process are also a function of the slurry particulate properties and the operational download. In the final part of this investigation, effects of slurry particle size and solids concentration on the performance of oxide CMP operations were investigated by examining material removal rate and surface quality responses as a function of applied load. The possible variations in the mechanical and chemical interactions were discussed as the slurry particulate properties varied, which underlined the importance of the formation of chemically modified surface layers during CMP operations. These layers were suggested to provide acceptable surface finish in addition to enabling the material removal in CMP. In silica polishing, presence of abrasives was observed to be required to obtain material removal, indicating that they were responsible for both the chemical modification of the surface and mechanical abrasion. Therefore, the particle-substrate interactions were investigated in detail for silica polishing. The slurry abrasives were detected to indent on the pad and the wafer surfaces at low solids concentrations leading to mechanical abrasion of the surfaces as they slide on the wafer. At higher solids concentrations, however, they started rolling since the load per particle was lower as compared to load applied per abrasive for less concentrated slurries. Although, rolling motion resulted in better surface finish, the higher material removal rates were obtained when the particles were sliding on the surfaces. The optimization analysis concluded that the most desirable polishing results are to be achieved with small particle size slurries at medium solids concentrations and under medium applied load conditions. The overall findings of this investigation led to the design criteria for the CMP slurries, which can be defined as follows: (i) presence of both hard and soft agglomerates must be avoided in
128 the slurries, (ii) robust stabilization schemes must be implemented by controlling the particle-particle and pad-particle-substrate interactions and (iii) monosized slurries with small particle size and medium solids concentrations should be used for polishing at medium operational pressures. Suggestions for Future Work The findings reported in this study can be further expanded to investigate several other issues, which are critical to increased performance of the CMP process. One of the main challenges faced in CMP is the development of effective predictive methodologies, which will lead to better control of CMP by enabling selection of effective operational and slurry properties. The predictive methodologies must account for the pad-particle-substrate interactions rather than the operational variables alone, as has been the common practice so far. In addition, the chemical effects must be taken into account to develop effective predictive methodologies, which lead to the formation of a chemically altered layer on the wafer surface during polishing. The mechanical properties of the chemically modified surface are believed to be very different than the substrate material to be polished. Therefore, unless these properties are investigated, the modeling efforts will be incomplete. The most effective technique to measure the mechanical properties of the chemically modified layers is nano-indentation analysis. However, they cannot provide the required sensitivity for the very thin films formed during CMP. The new techniques, such as ultrasonic force microscopy, introduced for the measurement of the mechanical properties of the ultra-thin films are more promising as they provide highly sensitive and nondestructive methodologies [Cro00, Yar00].
129 It has been verified in this investigation that the frictional forces play an important role on material removal during the CMP applications. Addition of surfactants, surfactant chain length, change in slurry ionic strength and valency of the added salt were observed to alter the lateral (frictional) interactions in polishing. Although a correlation was detected between the material removal rate and the measured frictional forces, the entire relationship is not known. Therefore, it is also required to model the impact of frictional interactions during the CMP applications. Furthermore, manipulation of the frictional forces during polishing by controlling surfactant adsorption/desorption can be used to design material selective slurries. Enabling the selective adsorption of surfactant molecules to a selected material while the other materials on the substrate remains uncoated, particle-substrate engagement could be achieved for the bare material leading to its polishing, whereas the coated material cannot be removed due to the presence of the lubricating layer. Naturally, these slurries must be tested with the patterned wafers to confirm the degree of selectivity they will provide during polishing. One final suggestion for future work relates to the slurry abrasive particle shape. Throughout this investigation, spherical particles were used for polishing. However, some of the commercial polishing slurries are made of fused silica, in which the primary particles are composed of chains of much smaller particle beads. These types of slurries are known to polish effectively. The shape of abrasives may alter the particle-substrate interactions as the particles tend to orient according to their shape during polishing. This can be utilized to modify the load applied per particle during CMP. Increase in the particle-substrate contact area will lead to reduction of the pressure created on the surface. While the spherical particles result in point contacts and relatively high pressure
130 generation, a plate like particle can generate much lower pressures, which are required for the polishing of softer and more sensitive materials such as copper and next generation polymeric dielectrics. Therefore, the impact of slurry particle shape on CMP performance is suggested to be studied as a part of future work for designing next generation CMP slurries.
APPENDIX CENTRAL COMPOSITE DESIGN AND RESPONSES Experimental variables: A: Particle Size (m) B: Slurry solids concentration (%wt) C: Applied head load (N) Responses: Material removal rate (MRR) (/min) RMS surface roughness (nm) Maximum depth of surface scratches or pits (Rmax) (nm) Final equations (* Significant factors in 90% confidence interval): Material removal rate = 4140*â€“ 6796 A*â€“ 189 B + 11 C* + 4286 A 2 *+ 5 B 2 *+ 0.06 C 2 + 89 A B * â€“ 16 A C â€“ 0.45 B C RMS Surface Roughness = 4.38* + 15.11 A *+ 0.23 B *â€“ 5.21 C â€“ 9.07 A 2 *â€“ 4.97 B 2 *+ 2.74 C 2 * â€“ 0.09 A B * â€“ 0.02 A C â€“ 2.98 B C Maximum depth of surface scratches or pits = 75.1 *+ 159.47 A *+ 0.15 B â€“ 3.1 C *â€“ 42.7 A 2 *+ 0.03 B 2 *+ 0.02 C 2 * â€“ 8.07 A B *+ 1.79 A C *+ 0.03 B C* 131
132 Run No Pattern Particle Size (m) Applied load (N) Solids Conc. (%wt) MRR (/min) RMS Roughness (nm) Rmax (nm) 1 -00 Axial 0.2 74 15 1620 0.68 20 2 0+0 Axial 0.5 114 15 992 4.23 109 3 000 Center 0.5 74 15 824 2.64 55 4 -++ Face 0.3 94 24 1688 1.86 53 5 +-Face 0.8 54 6 874 2.17 103 6 --+ Face 0.3 54 24 1463 1.02 22 7 000 Center 0.5 74 15 837 2.57 53 8 ++Face 0.8 94 6 1063 2.92 151 9 -+Face 0.3 94 6 2770 1.53 24 10 +-+ Face 0.8 54 24 868 2.03 40 11 000 Center 0.5 74 15 818 2.74 55 12 --Face 0.3 54 6 2070 0.79 25 13 00Axial 0.5 74 0.5 490 (outlier) 1.27 81 14 00+ Axial 0.5 74 30 1002 2.16 46 15 +00 Axial 1.0 74 15 682 2.39 107 16 000 Center 0.5 74 15 828 2.68 53 17 0-0 Axial 0.5 34 15 668 2.15 55 18 000 Center 0.5 74 15 813 2.67 53 19 +++ Face 0.8 94 24 902 2.25 93 20 000 Center 0.5 74 15 820 2.71 49
LIST OF REFERENCES Adl00 J. J. Adler, P.K. Singh, A. Patist, Y.I. Rabinovich, D.O. Shah and B.M. Moudgil, Langmuir, 16, 7255 (2000). Adl01 J. J. Adler, Interaction of Non-Ideal Surfaces in particulate Systems, Ph.D. Dissertation, University of Florida, Gainesville, FL (2001). Aga00 L. Agarwal and C-W Park, in Chemical Mechanical Polishing 2000Fundamentals and Materials Issues, R.K. Singh, R. Bajaj, M. Meuris, M. Moinpour and A. Maury, Editors, 613, p. 45, Materials Research Society Proceedings San Francisco, CA (2000). Aks96 I. A. Aksay, M. Trau, S. Manne, I. Honma, N. Yao, L. Zhou, P. Fenter, P. M. Eisenberger and S. M. Gruner, Science, 273, 892 (1996). Ask86 T. C. Askwith, A. Cameron and R. F. Crouch, Proc. Roy. Soc. A, 281, 500 (1986). Bas00 G. B. Basim, J.J. Adler, U. Mahajan, R. K. Singh and B. M. Moudgil, J. Electrochem. Soc., 147, 3523 (2000). Bas02 G.B. Basim and B.M. Moudgil, â€œEffect of Soft Agglomerates on CMP Slurry Performance,â€ J.Colloid Interface Science, Forthcoming, (2002). Bhu95a B. Bhushan, in Handbook of Micro/Nanotribology, B. Bhushan, Editor, p. 357, CRC Press, Boca Raton (1995). Bhu95b B. Bhushan, J. Israelachvili and U. Landman, Nature, 374, 607 (1995) Bie98 M. Bielmann, Chemical Mechanical Polishing of Tungsten, Masterâ€™s Thesis, University of Florida, Gainesville, FL (1998). Bie99 M. Bielman, U. Mahajan, R.K. Singh, D.O. Shah and B.J. Palla, Electrochem. Solid St., 2, 148 (1999). Bra00 A. E. Braun, Semiconductor International, 23 (10), 55 (2000). Bra01 A. E. Braun, Semiconductor International, 24 (11), 54 (2001). Bra02 A. E. Braun, Semiconductor International, 25 (4), 44 (2002). 133
134 Bre99 K. E. Bremmell, G. J. Jameson and S. Biggs, Colloid Surface A, 146, 75 (1999). Bro81 N.J. Brown, P.C. Baker and R.T. Maney, Proc. SPIE, 306, 42 (1981). Bud61 S. M. Budd, Phys. Chem. Glasses, 2, 111 (1961). Cab00 Cabot Corporation Microelectronics Division, Material Safety Data Sheet for Semi-Sperse 12 and 25 Aqueous Dispersions, Aurora, IL (2000). Car90 J.W. Carr, US Patent No. 4,954,142 (1990). Che92 Y. L. Chen, S. Chen, C. Frank and J. Israelachvili, J. Colloid Interf. Sci., 153, 244 (1992). Cla68 S. W. Clark and S. R. B. Cooke, Trans. AIME, 241, 334 (1968). Cle99 S. C. Clear and P.F. Nealey, J. Colloid Interf. Sci., 213, 238 (1999). Coh90 S. L. Cohen, V. A. Brusic, F. B. Kaufman, G. S. Frankel, S. Motake and B. Rush, J. Vac. Sci. Tech., A 8, 2417 (1990). Col97 M. Colic and D. W. Fuerstenau, Langmuir, 13, 6644 (1996). Coo90 L.M. Cook, J. Non-Cryst. Solids, 120, 152 (1990). Cop00 J. Coppeta, C. Rogers, L. Racz, A. Philipossian and F. B. Kaufman, J. Electrochem. Soc., 147, 1903 (2000). Cor63 D. Cornish and L. Watt, Br. Sci. Instr. Res. Assoc. Rep., R295 (1963). Cro00 K. B. Crozier, G. G. Yaralioglu, F. L. Degertekin, J.D. Adams, S. C. Minne and C.F. Quate, Appl. Phys. Lett., 76, 1950 (2000). Dor73 R. Doremus, Glass Science, Wiley, New York, NY (1973). Dov92 P. M. Dove and S. F. Elston, Geochim. Cosmochim. Acta, 56, 4147 (1992). Duc92 W. A. Duccer, T. J. Senden and R. M. Pasheley, Langmuir, 8, 1831 (1992). Duk98 A.S. Dukhin and P.J. Goetz, Colloid Surface A, 144, 49 (1998). Eva96 C. R. Evanko, D. A. Dzombak and J. W. Novak, Colloids Surface A, 110, 219 (1996).
135 Ewa99 R. Ewasiuk, S. Hong and V. Desai, in Chemical Mechanical Polishing in IC Device manufacturing III, Y.A. Arimoto, R.L. Opila, J.R. Simpson, K.B. Sundaram, I. Ali, Y. Homma, Editors, p. 408, 99-37, Electrochemical Society Proceedings Series, Pennington, NJ (1999). Fle98 T.D. Fletcher, L.C. Hardy, D. Kaisaki, H. Kranz, J. Trice and R. Webb, 3-M Company, CAMP 3 rd Annual International Symposium on CMP, Lake Placid, NY (1998). Fue95 M. C. Fuerstaneu, in Flotation Science and Engineering, K. A. Matis, Editor, p.89, Marcel Dekker Inc, New York, NY (1995). Hou92 W. A. House and D. R. Orr, J. Chem. Soc., Faraday Trans., 88, 233 (1992). Ile79 R. Iler, The Chemistry of Silica: Solubility, Polymerization, Colloid and Surface Properties, and Biochemistry, Wiley, New York, NY (1979). Isr92 J.N. Israelachvili, Intermolecular and Surface Forces, Academic Press, New York, NY (1992). Ito81 S. Ito and M. Tomozawa, J. Am. Ceram. Soc., 64, C-160 (1981). Izu79 T. Izumitani, in Treatise on Materials Science and Technology, M. Tomozawa, R. Doremus, Editors, p. 115, Academic Press, New York, NY (1979). Izu84 T. Izumitani, Paper TuB-A1, Tech. Digest, Topical Meeting on the Science of Polishing, Optical Society of America (1984). Jai94 R. Jairath, M. Desai, M. Stell, R. Tolles and D. Scherber-Brewer, in Advanced Metallization for Devices and Circuits-Science, Technology and Manufacturability, S.P Murarka, A. Katz, K.N. Tu and K. Maex, Editors, 337, p. 121, Materials Research Society Proceedings Series, Pittsburgh, PA (1994). Kau91 F. B. Kaufman, D.B. Thomson, R.E. Broadie, M.A. Jaso, W.L. Guthrie, D. J. Pearson and M. B. Small, J. Electrochem. Soc., 138, 3460 (1991). Kir94 N.B. Kirk and J.V. Wood, British Ceramic Trans., 93 (1), 25 (1994). Kle94 J. Klein, E. Kumacheva, D. Mahalu, D. Perahia and L. J. Fetters, Nature, 370, 634 (1994). Koo99 L. K. Koopal, T. Goloub, A. de Keizer and M. P. Sidorova, Colloid Surface A, 151, 15 (1999). Kum90 K. Kumar, D. O. Shah, in Tribology and the Liquid-Crystalline State, G. Biresaw, Editor, 441, p. 91, ACS Symposium Series, Washington, DC (1990).
136 Lan85 W. Lanford, C. Burman and R. Doremus, in Advances in Materials Characterization II, R. Snyder, R. Condrate and P. Jonston, Plenum, Editors, p. 205, New York, NY (1985). Lan92 H. Landis, P. Burke, W. Cote, W. Hill, C. Hoffman, C. Kaanta, C. Koburger, W. Lange and S. Luce, Thin Solid Films, 347, 248 (1992). Lea97 H. J. Leamy and J. H. Wernick, MRS Bull., 47 (5) (1997). Li94 J. Li, T. E. Seidel and J. W. Mayer, MRS Bull., 19 (8), 15 (1994). Luo98 Q. Luo, S. Ramajan and S.V. Babu, Thin Solid Films, 335, 160 (1998). Man94 S. Manne, J. P. Cleveland, H. E. Gaub, G. D. Stucky and P. K. Hansma, Langmuir, 10, 4409 (1994). Man95 S. Manne and H. E. Gaub, Science, 270, 1480 (1995). Mah99a U. Mahajan, M. Bielman and R.K. Singh, Electrochem. Solid St.., 2, 46 (1999). Mah99b U. Mahajan, M. Bielmann and R.K. Singh, Electrochem. Solid St., 2, 80 (1999). Mah00 U. Mahajan, Fundamental Studies on Silicon Dioxide Chemical Mechanical Polishing, Ph.D Dissertation, University of Florida, Gainesville, FL (2000). Mas89 R. L. Mason, R. F. Gunst and J. L. Hess, Statistical Design and Analysis of Experiments with Applications to Engineering and Science, John Wiley & Sons, New York, NY (1989). Mat96 S. Mathur, Adsorption Mechanism(s) of Poly(ethylene oxide) on Oxide and Silicate Surfaces, Ph.D. Dissertation, University of Florida, Gainesville, FL (1996). Mau97 A. Maury, D. Ouma, D, Bonning and J. Chung, Conference for Advanced Metallization and Interconnect Systems for ULSI Applications, San Diago, CA (1997). McB86 J. D. McBrayer, R. M. Swanson and T. W. Sigmon, J. Electrochem. Soc., 133, 1243 (1986). Mic82 T. A. Michalske and S. W. Freiman, Nature, 295, 511 (1982). Moy89 D. Moy, M. Schadt, C.K. Hu, F. Kaufman, A.K. Ray, N. Mazzeo, E. Baran and D.J. Pearson, Proc. 6 th VMIC, Santa Clara, CA, 26 (1989).
137 Mur91 S. P. Murarka, in Tungsten and Other Advanced Metals for VLSI Applications in 1990, G. C. Smith and R. Blumenthal, Editors, 179, p. 89, Materials Research Society Proceedings Series, Pittsburgh, PA (1991). Mur93 S. P. Murarka, Metallization Theory and Practice for VLSI and ULSI, Butterworth-Heinemann, Boston, MA (1993). Mur00 S. P. Murarka, in Chemical-Mechanical Polishingâ€”Fundamentals and Challenges, S. V. Babu, S. Danyluk, M. I. Krishnan and M. Tsujimura, Editors, 566, p. 3, Materials Research Society Proceedings Series, Warrendale, PA (2000). Nag96 R. Nagahara, S.K. Lee and H.M. You, Particle Sizing Systems, VMIC Conference, Jun. 18-20, 1 (1996). Ngu01 V. H. Nguyen, A.J. Hof, H. van Kranenburg, P. H. Woerlee and F. Weimar, Microelectronic Engineering, 55, 305 (2001). Nog84 M. Nogami and M. Tomozawa, Phys. Chem. Glasses, 25, 82 (1984). Obe98 Y. Obeng, K. M. Forsthoefel, k. A. Richardson and R. H. Burton in Chemical Mechanical Polishing in IC Device Manufacturing II, I. Ali, S. Raghavan, and R. L. Opila, Editors, 98-7, p. 151, Electrochemical Society Proceedings Series, Pennington, NJ (1998). Oer85 G. Oertel, Polyurethane Handbook, Macmillan Pub. Co, New York, NY (1985). Ols93 J. Olsen and F. Moghadam, in Multilevel Metallization for Integrated Circuits, S.R. Wilson, C.J. Tracy and J.L. Freeman, Editors, Noyes Publications, Park Ridge, NJ (1993). Oum99 D.O. Ouma, Modelling of Chemical Mechanical Polishing for Dielectric Planarization, PhD Dissertation, MIT, Boston, MA (1999). Pet02a L. Peters, Semiconductor International, 25 (1), 48 (2002). Pet02b L. Peters, Semiconductor International, 25 (5), 55 (2002). Pal00 B. J. Palla and D. O. Shah, J. Colloid Interf. Sci., 223, 102 (2000). Poh96 M. C. Pohl and D. A. Griffiths, J. Electron. Mater., 10, 1612 (1996). Pre27 F. Preston, J. Soc. Glass Technol., 11, 247 (1927).
138 Qin03 K. Qin M., A Modelling Process for Chemical Mechanical Polishing, Forthcoming Ph.D. Dissertation, University of Florida, Gainesville, FL (2003). Raj97 K. Rajan, R. Sigh, J.J. Adler, Y. Rabinovich and B.M. Moudgil, Thin Solid Films, 308, 529 (1997). Ram00 S. Ramajan, Y. Li, M. Hariharaputhiran, Y. S. Her and S. V. Babu, in Chemical Mechanical Polishing â€“ Fundamentals and Challenges, S.V. Babu, S. Danyluk, M. Krishnan and M. Tsujimura, Editors, 566, p. 97, Materials Research Society Proceedings Series, Pittsburgh, PA (2000). Run94a S. R. Runnels and L.M. Eyman, J. Electrochem. Soc., 141, 1698 (1994). Run94b S. R. Runnels, J. Electrochem. Soc., 141, 1900 (1994). Run96 S. R. Runnels, J. Electron. Mater., 25, 1574 (1996). Sch95 J. Schlueter, SemiCon Southwest, 95, 1 (1995). Shi98 F.G. Shi, B. Zhao and S.-Q. Wang, Proc. of International Technology Conference, San Francisco, CA, 73 (1998). Sin01 P. K. Singh, J. J. Adler, Y. I. Rabinovich and B. M. Moudgil, Langmuir, 17, 468 (2001). Sin02 P. Singer, Semiconductor International, 25 (5), 46 (2002). Siv92 S. Sivaram, M. H.M. Bath, E. Lee, R. Leggett and R. Tolles, Proc. SRC Topical Research Conference on Chem-Mechanical Polishing for Planarization, SRC, Research Triangle Park, NC, 8 (1992). Sjo96 S. Sjoberg, J. Non-Cryst. Solids, 196, 51 (1996). Sol99 M. J. Solomon, T. Saeki, M. Wan, P. J. Scales, D. V. Boger and H. Usui, Langmuir, 15, 20 (1999). Ste94 J.M. Steigerwald, S.P. Murarka, D.J. Duquette, R.J. Gutmann, in Advanced Metallization for Devices and Circuits-Science, Technology and Manufacturability, S. P. Murarka, A. Katz, K. N. Tu, and K. Maex, Editors, 337, p.133, Materials Research Society Proceedings Series, Pittsburgh, PA (1994). Ste97 J.M. Steigerwald, S.P. Murarka and R.J. Gutmann, Chemical Mechanical Planarization of Microelectronic Materials, John Wiley and Sons, New York, NY (1997).
139 Ste99 D. J. Stein, D. L. Hetherington and J. L. Cecchi, J. Electrochem. Soc., 145, 1934 (1999). Sti97 G. Stix, Sci. Am., 276, 45 (1997). Sze81 S. M. Sze, Physics of Semiconductor Devices, John Wiley Sons Inc., New York, NY (1981). Tho94 M. E. Thomas and I. Saadat, Topical Research Conference on Low Dielectric Constant Interlayer Dielectics for High Performance Circuits, SRC, Research Triangle Park, NC (1994). Tic99 J. Ticky, J. A. Levert, L. Shan and S. Danyluk, J. Electrochem. Soc., 146, 1523 (1999). Tom94 M. Tomozawa, K. Yang, H. Li and S.P. Murarka, in Advanced Metallization for Devices and Circuits-Science, Technology and Manufacturability, S.P Murarka, A. Katz, K.N. Tu and K. Maex, Editors, 337, p. 89, Materials Research Society Proceedings Series, Pittsburgh, PA (1994). Tri99 J. Trice, T.D. Fletcher, L.C. Hardy and J. Kollodge, 3-M Company, CAMP 3 rd Annual International Symposium on CMP, Lake Placid, NY (1999). Tro94 J. A. Trogolo and K. Rajan, J. Mater. Sci., 29, 4554 (1994). Tse97 W-T. Tseng and Y-L. Wang, J. Electrochem. Soc., 144, 14 (1997). Utt91 R. R. Uttecht and R. M. Geffken, Proc. 8 th VMIC, Santa Clara, CA, 20 (1991). Vel00 P. Velden, Microelectron. Eng., 50, 41 (2000). Vog99 W. Vogelsberger, M. Lobbus, J. Sonnefeld and A. Seidel, Colloid Surface A, 159, 311 (1999). Wan94 S. Q. Wang, MRS Bull., 19 (8), 15 (1994). Wol86 S. Wolf and R. N. Tauber, Silicon Processing for the VLSI Era, Lattice Press, Sunset Beach, CA (1986). Xia96 X. Xiao, J. Hu, D. H. Charych and M. Salmeron, Langmuir, 12, 235 (1996). Xie96 Y. Xie and B. Bhushan, Wear, 200, 281 (1996). Xu00 C. Xu and W. Gao, Mat. Res. Innovat., 3, 231 (2000).
140 Yar00 G. G. Yaralioglu, F. L. Degertekin, K. B. Croizer and C. F. Quate, J. Appl. Phys., 87 (10), 1 (2000). Ye93 H. Ye, I. Matsuoka, Int. J. Miner. Process., 40, 123 (1993). Yu94 T.K. Yu, C.C. Yu and M. Orlowski, Proceedings of the 1993 International Electronic Devices Meeting, 4.1, 35 (1994). Zha00 G. Zhang, Semiconductor International, 23 (6), 40 (2000).
BIOGRAPHICAL SKETCH Gul Bahar Basim was born on May 4, 1973, in Ankara, Turkey. She graduated first in the graduating class in 1990 from Karsiyaka Gazi High Scholl in Izmir, Turkey. She then enrolled at Middle East Technical University in Ankara, Turkey. In July 1995 she graduated first in the graduating class with a Bachelor of Science degree in mining engineering with a specialization in minerals processing. She started her graduate study at Virginia Polytechnic Institute and State University in 1996, where she pursued a Masters of Science degree in mining and minerals engineering. Her masterâ€™s thesis work was supported by the Department of Energy and concentrated on fine coal dewatering under the advisory of Professor Roe-Hoan Yoon. In 1998, she has started her Ph.D. with Professor Brij Moudgil at the University of Florida, where she has worked as a part of the Engineering Research Center for Particle Science and Technology. Her dissertation research concentrated on formulation of chemical mechanical polishing slurries. She graduated from University of Florida with a doctorate degree in materials science and engineering with electronic materials and particle science and technology specialties in August 2002. 141