|
Citation |
- Permanent Link:
- https://ufdc.ufl.edu/UFE0012201/00001
Material Information
- Title:
- Particle Scale Modeling of Material Removal and Surface Roughness in Chemical Mechanical Polishing
- Creator:
- YERUVA, SURESH BABU ( Author, Primary )
- Copyright Date:
- 2008
Subjects
- Subjects / Keywords:
- Chemicals ( jstor )
Hardness ( jstor ) Mechanistic materialism ( jstor ) Modeling ( jstor ) Particle size classes ( jstor ) Particle size distribution ( jstor ) Polishing ( jstor ) Semiconductor wafers ( jstor ) Slurries ( jstor ) Surface layers ( jstor )
Record Information
- Source Institution:
- University of Florida
- Holding Location:
- University of Florida
- Rights Management:
- Copyright Suresh Babu Yeruva. Permission granted to University of Florida to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
- Embargo Date:
- 7/30/2007
- Resource Identifier:
- 76786526 ( OCLC )
|
Downloads |
This item has the following downloads:
|
Full Text |
PARTICLE SCALE MODELING OF MATERIAL REMOVAL AND SURFACE
ROUGHNESS IN CHEMICAL MECHANICAL POLISHING
By
SURESH BABU YERUVA
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
2005
Copyright 2005
by
Suresh Babu Yeruva
This document is dedicated to Sri Paramahansa Yogananda, my parents and sister.
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 towards his students encouraged me to perform my best and expand my
potential. Sincere thanks are also due to Dr. Chang-Won Park, for his valuable guidance
and support throughout the study. I would also like to acknowledge the other members of
my advisory committee, Dr. Rajiv Singh, Dr. Wolfgang Sigmund, Dr. Hassan El-Shall,
Dr. Susan Sinnott and Dr. Dinesh Shah, for their time, helpful discussions and guidance.
I would like to thank Dr. Yakov Rabinovich for help with atomic force microscope
(AFM) and Fourier transform infrared spectroscopy (FTIR) analysis, and Gerald Bourne
from Major Analytical Instrumentation Center (MAIC) for help with focused ion beam
(FIB) and nanoindentation data. Furthermore, I sincerely acknowledge the guidance and
support given by Kuide Qin at Praxair Inc.
I gratefully acknowledge the National Science Foundation's Particle Engineering
Research Center (PERC) for financially supporting this research. Special thanks are due
to PERC staff members, Gary Schieffele, Gill Brubaker, Mike Beasley, and John
Henderson who made this work possible.
My colleagues, Bahar Basim, Jeremiah Abiade, Su-Ho Jung, Vijay Krishna,
Madhavan Esayanur, Scott Brown, Parvesh Sharma, Kyoung-Ho Bu, Pankaj Singh,
Debamitra Dutta and Ivan Vakarelski, who have contributed to this research by their
valuable discussions and friendship, are also gratefully acknowledged.
TABLE OF CONTENTS
Page
L IS T O F T A B L E S .............................................. ................................... .................. v ii
LIST OF FIGURES ................... ...................... .. ................... viii
ABSTRACT xi
CHAPTER
1 IN TR OD U CTION ............................................... .. ......................... ..
M ultilevel M etallization ................................................................ ...................... 2
N eed fo r P lan arizatio n ............................................ ................................ ................. ........ 4
Planarization Techniques.................... ......... ............... ........ .......................
Overview of the Chemical Mechanical Polishing Process............... ..................7
H history ................... .. .. ..... .. ... ................ ... ... ............................ . 7
Advantages and Disadvantages of Chemical Mechanical Planarization............8
G general C M P applications.......................................................... ............... 9
C m p O utput Param eters ............................................................................ 9
R em ov al rate .............................................................10
Global planarization ............... .. .................... ........ .............. .. 12
S ele ctiv ity ................................ ...................1.........3
M easurem ent T techniques .......................................... ................. ... ............ 14
End Point D election ............................... .............. ......... ........ .... 15
CMP Consumables ............................... .......... ...... ............... 16
C M P S lu rry ............... ............................ ............................... 1 8
R ate of surface reaction ..................................................... ..................... 18
Slurry agglomeration...................... ....... ............................ 19
Chemical M echanical Polishing Pads ...................................... ............... 20
Effect of pad grooves ................................................. .......... .................. 23
Physical and chemical changes in pad during polishing ............................23
Effect of temperature change on polishing pad ..........................................26
Non-uniformity of the polishing pad............................ .................27
C M P process in the future ............................................................... .....................27
S cop e of D issertation ......................................................................... ...................2 8
2 CMP MODELING: OVERVIEW OF PERC I................................... ...............30
L iteratu re R ev iew ....................... ........................................ ........................ .. .3 1
v
Preston Equation ............................................. ........ ... .......... .... 31
Modifications to Preston Equation ........................................ ...............34
Slurry Erosion............................................. 36
Pad Considerations in CM P M odels......................................... ............... 39
Slurry Considerations in CM P M odels..................................... ............... 42
PER C I CM P M odel ........................ .......... ....... .............................44
M odel D evelopm ent ........... ................ .................... .................. 45
Estimation of the Overall Removal Rate.................................54
3 PER C II CM P M O D EL ............... ................................................... ............... 59
A ssum option V alidation ......... ........... ............... ....................... ...............59
P ad-w afer C contact A rea ........................................................... .....................59
Optical m icroscopy ................. .......................................... 61
Fourier transform infrared spectroscopy (FTIR).........................................64
Pad Asperity Height Distribution ................... .................... ...............73
Abrasive Particle Size D distribution .......................................... ............... 74
PER C II M odel D evelopm ent.......................................................... ............... 75
Plastic Deformation of Wafer Surface .......................................... ..........79
Slurry Characteristics Particle Size Distribution.................. ..................79
M material Rem oval by Single Particle........................................ ............... 79
Number of Active Abrasive Particles ................... .....................82
Estimation of the Overall Removal Rate ..................................... ...............83
4 M O D E L A SSE SSM E N T ......................................... .............................................85
M odel Validation .............................. ....... .................................85
Silica C M P : M icron Sized Slurries ........................................ .....................86
Silica CM P: N anosilica slurries...................................... ......................... 87
Copper CMP ............. .. ........ ............. ......... 91
Tungsten CM P ......... ... .............. .. ....................... .... ........92
M odel P reductions ......... .. ............. .................. ............... .. .. ............ ......... ..........94
Removal Rate as a Function of Particle Size for a Given Solids Loading:.........95
Effect of Modified Layer Thickness ............... ......................................96
Effect of M odified Layer Hardness .............. .............................................. 99
5 SUMMARY AND SUGGESTIONS FOR FUTURE WORK ........... .....................101
L IST O F R E F E R E N C E S ...................................................................... ..................... 106
BIOGRAPHICAL SKETCH ........................................................................118
LIST OF TABLES
Table p
1-1. Interconnection delay (RC) in silicon VLSI chip............................................ ............ 3
1-2. Chemical mechanical polishing applications. ................................. .................10
4-1. Parameters included in the revised PERC model................................................. 85
4-2. Parameter values used for comparing PERC I and II models. ..................................95
5-1. Process parameter ranges used in sensitivity analysis ................ ......... ..........103
LIST OF FIGURES
Figure pge
1-1. Future trends in interconnects .................................. ...................................... 4
1-2. Schematic of a device fabricated using multilevel metallization. ..............................5
1-3. Different degrees of local and global surface planarity..................... .....................6
1-4. Schematic of chemical mechanical polishing process.............. ................................. 7
1-5. Surface layer formation on Cu with various slurry chemistries. ...........................11
1-6. M icro and nanoscale phenomena during CM P ....................................................13
1-7. Parameters involved in CMP process................................ ...............17
1-8. Chronoamperometry measurements for tungsten ............................................... 19
1-9. Type I C M P pad.. ................. ...... .......................... .. ...... .. ............. 22
1-10. Type II CM P pad. ............................................. .. .... ........ ......... 22
1-11 The absorbance spectra of used and dry IC 1000 pad along with dried silica
slu rry ................................................... ...................... ................ 2 5
2-1. Polished wafer surfaces showing nanoscratch surfaces. .........................................33
2-2 Pad consideration in PER C I. ........................................................... .....................46
2-3. Schematic of single particle abrasion showing the removal of chemically
m modified surface layer............ .. .................. ........ ...... .. .. ........ .... 51
2-4. Schematic of PERC I predictions for increasing particle size..............................57
2-5. Polishing of silica by nanoslurry showing optimal particle size.. ............................57
2-6. Polishing of silica at various pressure, velocity and particle sizes showing optimal
p article siz e .........................................................................5 8
3-1. Schematic of the fixture used in measuring pad-wafer (dry) contact area. ................62
3-2. Pad asperity contact spots seen through optical microscope................... ................63
3-3. Pad asperity contact area dependence on applied load. Each legend represents a
selected asperity contact .................. ........................... ....... .... ........ .... 63
3-4. Pad contact area as a function of applied load................................. ............... 64
3-5 Schematic of the FTIR-ATR technique used to measure pad-wafer contact area.......65
3-6. Analysis of the FTIR-A TR spectra................................................. ....... ........ 67
3-7. Absorbance spectra of pad (in pH 4) at different applied pressures...........................67
3-8. Contact area variation with applied load in dry and wet conditions ........................68
3-9. Contact area in dry and wet systems in various pH conditions................................68
3-10. C olloidal probe A FM tip ................................................ .............................. 70
3-11. Force-distance curve indicating softer "modified layer" on the pad surface. ..........71
3-12. Force-indentation curves from AFM for different conditions...............................72
3-13. Schematic of pad surface as deduced from AFM data. ............. ..............72
3-14. Schem atic of polishing pads. ........................................ ......................................73
3-15. Particle size distribution of silica slurry. ...................................... ............... 75
3-16. Schematic of pad, slurry and wafer in PERC I model.................... ..................76
3-17. Schematic of pad, slurry and wafer in PERC II model. ........................................77
4-1. Slurries used in oxide polishing................. ........................................ 87
4-2. Silica polishing by slurries in the micron range. .............................. ......... ...... .88
4-3. Silica polishing by colloidal nanosilica. .............................................................. 89
4-4. Silica polishing by nanosized slurries ...................................................................90
4-5 Copper CMP as a function of particle size...............................................................92
4-6. Alumina slurry size distributions..................... ...... ............................. 93
4-7. Effect of particle size on tungsten CMP .............. ......... ......... ...............94
4-8. Comparison between PERC I and PERC II models with constant solids loading of
different particle sizes.. ....................................... ................. ......... 96
4-9. Model comparison with varying modified layer thickness.. ...................................98
4-10. CMP performance comparison with varying modified layer hardness ................100
5-1. Sensitivity analysis of silica CM P system.............. .............................................. 103
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
PARTICLE SCALE MODELING OF MATERIAL REMOVAL AND SURFACE
ROUGHNESS IN CHEMICAL MECHANICAL POLISHING
By
Suresh Babu Yeruva
December 2005
Chair: Brij M. Moudgil
Cochair: Chang-Won Park
Major Department: Materials Science and Engineering
Chemical mechanical polishing (CMP) is widely adopted in producing excellent
local and global planarization of microelectronic devices. It has been demonstrated
experimentally that the polishing performance is a result of the synergistic effect of both
the chemicals and the particles involved in CMP. However, the fundamental mechanisms
of material removal and the interactions of the chemical and mechanical effects are not
well understood. A comprehensive model for CMP was developed taking into account
both the chemical and mechanical effects for slurries with a given particle size
distribution. The model developed (PERC II) is based on a previously developed model
(PERC I). The chemical aspect is attributed to the chemical modification of the surface
layer due to slurry chemistry, whereas the mechanical aspect is incorporated by
indentation of particles into the modified layer and the substrate depending on the
operating conditions. In this study, the effects of particle size and pad asperity
distributions are included in the model.
The contact area of pad with wafer was measured in dry and wet conditions in
different pH solutions using optical microscopy and Fourier transform infrared
spectroscopy respectively. Pad surface mechanical properties in dry and wet states were
also investigated using atomic force microscopy. The contact area results obtained were
utilized in modeling to estimate the pad modulus leading to pad-wafer contact stress
distribution.
The predictions of the model show a reasonable agreement with the experimental
data. The model is validated for oxide and metal CMP systems. The PERC II model not
only predicts the overall removal rate, but also the surface roughness of the polished
wafer in selected systems. The developed model can be used to optimize the current CMP
systems and provide insights into future CMP endeavors.
CHAPTER 1
INTRODUCTION
The past three decades have witnessed rapid growth of the electronics industry and
its impact on human lifestyle. Ultra-large-scale integrated (ULSI) circuits, with 108 or
more devices on a chip, can now be fabricated on semiconductor wafers, reducing cost
and increasing the performance of electronic products. The number of components on a
memory chip has approximately doubled every two years over the past two decades,
following the famous Moore's law [1]. The advancement in fabrication technology
shrinks the minimum dimension of the device-feature by about 13% per year. This trend
is expected to continue in the future even as 65 and 45 nm processes are set for
production in 2005 [2] and near future.
The relentless competitor and customer driven demand for increased circuit
density, functionality and versatility has led to revolutionary advances in the "front end"
of the chip manufacturing line where the circuit elements are fabricated, and the "back
end" where these elements are appropriately wired within the integrated circuit (IC) [3].
Chip interconnections, or "interconnects," serve as local and global wiring, connecting
circuit elements and distributing power [4]. To incorporate and accommodate the
improvements such as decreased feature size, increased device speed and more intricate
designs, research in the back end of line (BEOL) processes has become equally important
as the development of the front end of line (FEOL) processes to reduce gate oxide
thickness and channel length.
While recent path breaking innovations [5-7] in the field of lithography and
patterning have brought progressive device scaling, the development of a planar BEOL
approach, which incorporates the use of chemical mechanical polishing (CMP) to
planarize inter-level dielectrics and metal stud levels, represents a significant
advancement in BEOL processes. Innovation in BEOL technology is required in each
technology generation, since only part of the density increase could be achieved with
improvements in lithography.
Multilevel Metallization
In earlier stages, the complementary metal oxide semiconductor (CMOS) device
structure had multiple isolated devices connected by a single level of interconnect wiring.
Scaling down of the devices was very effective in achieving the goals of increased device
density, functional complexity and performance. However, device scaling became less
profitable, and speed and complexity were dependant on the characteristics of
interconnects that wired them [8]. With the single level metallization scheme, the total
area occupied by the wiring on the chip significantly increased with the increase in the
active density on the chip.
The interconnect delay is usually measured as RC time delay, which is the total
time taken by the voltage at one end of the metal line to reach 63% of the total value of
the step input applied at the other end. This reduction is due to resistance of the
interconnect wiring metal (R) and the interlayer dielectric capacitance (C) [9]. The
resistance of the line is given by
R =p l/wd (1.1)
where p is the resistivity of the wiring material, 1, w, and d are the length, width, and
thickness of the wiring material respectively. The capacitance of the line is given by
C = s wl/t (1.2)
where s is the relative permittivity and t is the thickness of the insulator. The total RC
delay can be given by
RC= p 1/wd s wl/t =ps 12/td (1.3)
Thus, it can be seen from Eq. (1.3) that RC delay is independent of the line width
and further scaling of line width translates into reduction of IC line thickness which in
turn increases the RC delay. Other factors such as parasitic capacitances and cross-talk
become dominant for sub 0.5 |tm integrated circuits. One approach to decrease the RC
delay is by incorporating metals of low resistivity, and interlayer dielectrics of lower
dielectric constant. Table 1-1 gives the simple RC time constants calculated for a few
metals of given Rs (sheet resistance) and 1 mm length on 1 tm thick SiO2 [9]. The other
approach implemented widely in the industry is forming multilevel metallization schemes
where different levels of metal interconnections are isolated by dielectrics and are
connected by vertical vias.
Table 1-1. Interconnection delay (RC) in silicon VLSI chip.
Metal Bulk Polycrystalline Film Rs Delay'
resistivity film resistivity thickness (Q /square) (ps/mm)
(l2 cm) (S2 cm) (A)
Polysilicon -1000 5000 20 690
CoSi2 10 15 2500 0.6 21
MoSi2 -35 -100 2500 4 138
TaSi2 45 55 2500 2.2 76
TiSi2 13 15 2500 0.6 21
WSi2 -25 70 2500 2.8 97
W 5.65 8-10 2500 0.32-0.4 11-14
Mo 5.2 8-10 2500 0.32-0.4 11-14
Al 2.65 2.7 2500 0.11 4
Cu 1.67 2.0 2500 0.08 3
'Delay=RC=34.5 Rs (ps/mm) for 1 mm length conductor on l[tm thick SiO2 [10].
The increase in the levels of the metallization lines implies that packing density
need not keep pace with the device density and the minimum metal line feature does not
have to scale at the same pace as the gate width. The foremost reason behind the
implementation of multilevel metallization schemes is the reduction in the length of the
metal lines, thereby reducing the RC delay sizably (Eq. 1.3), and allowing direct routing
of the active devices. In places where metal wiring length cannot be reduced, routing can
be done at the upper levels without reducing the metal line width, thus reducing the RC
delay due to the higher surface area. Starting with two levels of metallization, the levels
of metallization have increased up to 8 by 2001 [11]. The future trends in the levels of
metallization can be seen in Fig. 1-1 [12].
--III ..."
Fig. 1-1. Future trends in interconnects [12]
Need for Planarization
Deposition of multiple layers leads to conformation of deposited layer on the
substrate features or already deposited pattern. When multilevel metallization is carried
out without any planarization step, non-planarized surface topography (Fig. 1-2a) results
in several processing difficulties. The irregular surface causes a hindrance in conformal
coating of the photoresist and efficient pattern transfer with contact lithography. The
anomalies in the surface cause the variation of the thickness in fine line widths (sub 0.5
[tm) depending on photoresist thickness. Fig. 1-2 shows a comparison between planarized
and non-planarized surface topography. Effectively planarized surfaces have a number of
benefits such as (a) higher photolithography and dry etch yields, (b) elimination of step
coverage concerns, (c) minimization of prior level defects, (d) elimination of contact
interruption, undesired contacts and electro-migration effects, (e) reduction of high
contact resistance and inhomogeneous metallization layer thickness, and (f) limitation in
the stacking height of metallization layers. Thus the demand for developing devices with
enhanced performance necessitates planarization.
Fig. 1-2. Schematic of a device fabricated using multilevel metallization. (a) Non-
planarized structure. (b) Planarized structure [8, 13]
Planarization Techniques
Techniques such a spin on deposition (SOD), reflow of boron phosphorous silicate
glass (BPSG), spin etch planarization (SEP), reactive ion etching and etch back (RIE
EB), SOD + EB and chemical mechanical polishing (or chemical mechanical
planarization) (CMP) are the most prominent of several competing technologies available
to achieve local and global planarization (Fig. 1-3). The plasma enhanced chemical vapor
deposited oxides have limited capability of gap filling and are restricted in their gap
filling ability below patterns having 0.3-|jm feature size. High-density plasma deposited
oxides have acceptable gap filling capabilities; however, they produce variation in
surface topography at local as well as global level. Even though spin-on deposited (SOD)
doped and undoped oxides and polymeric materials have acceptable ability for gap
filling, CMP is the only technique which produces excellent local and global
planarization of microelectronic materials.
(a) 1.nplanarizecd Gap Fill
-- IJ t YC D | 1
b1 Sur'ace Smoathig SOD (FSGTPSG)
(c) Locallianarizadtni SODWEB (BPSG)
Resist Eth Back
(d) GIobal PH uizatiou CMP
s 10pm
Fig. 1-3. Different degrees of local and global surface planarity [14].
Overview of the Chemical Mechanical Polishing Process
Current semiconductor fabrication technology for logic and memory devices
requires CMP to achieve the required multilevel interconnection densities. Indeed, each
silicon wafer can be exposed to 15 or more CMP steps before final device assembly. A
schematic diagram of the CMP process is shown in Fig. 1-4. In CMP, the wafer is
pressed face down against a rotating polishing pad, while a chemical and physically
abrasive slurry fed between pad and wafer planarizes the wafer. The material removal in
simplest terms takes place by synergistic mechanism between chemical reaction on the
wafer surface due to the slurry, and physical removal of the reacted material by the slurry
abrasives. The various mechanisms between pad, slurry and wafer are tailored to enhance
material removal, planarization, surface finish and higher selectivity compared to the
underneath layer.
Wafer Carrier Slrry,-"
..... Polish Pad
Platen
Fig. 1-4. Schematic of chemical mechanical polishing process.
History
The modern day application of the CMP process in the semiconductor industry was
for polishing the raw silicon wafers to achieve a global flatness. One of the most
widespread applications of CMP outside the semiconductor industry is in optical lens
polishing. In fact, the first semiconductor CMP machine was an innovation of the lens
polishing industry. The first wafers polished using CMP were commercially available in
the early 1960s [15]. There was wide spread industrial implementation of the different
variants of the CMP by companies such as NEC, National Semiconductor, Hitachi, etc.
[10]. This led to the introduction of the first commercial polisher designed specifically for
CMP by Cybeg, Japan in 1988. Later, International Sematech identified CMP as a
technology critical for the future of IC manufacturing and launched a project to develop
competitive and advanced CMP tools in the US [16]. Once the technology and required
equipment were available, CMP was used beyond polishing of inter-level dielectric layers
to polishing of tungsten, which was deposited in via openings between metal layers.
Copper was not used for interconnects until the availability of CMP even though it has
lower resistivity than aluminum [17], as it could not easily be etched by reactive ion
etching. This enabling polishing technology has spread to several other applications, the
significant one being in shallow trench isolation (STI) [18].
Advantages and Disadvantages of Chemical Mechanical Planarization
The advantages of CMP are (a) it achieves global planarization, (b) a wide range of
materials can be planarized, sometimes in the same polishing step, (c) it reduces severe
topography to allow fabrication with tighter design rules, (d) it provides an alternate
means of patterning metal, eliminating the need to plasma etch, (e) it improves metal step
coverage due to reduction in topography and (f) it does not use hazardous gasses common
in dry etch processes.
Conversely, the disadvantages of the CMP process are (a) poor control over the
process with narrow latitude in process variables, (b) new types of defects introduced by
the CMP process itself (scratches, particle residue, etc.), (c) difficult end-point detection
for desired thickness and (d) expensive to operate because of costly equipment and
consumables.
General CMP applications
The process of CMP was initially developed and implemented for planarization of
SiO2 which is used as an interlayer dielectric in a multilevel metallization scheme.
Tungsten is used as an interconnect plug to the source, drain, and gates of transistors in Si
microprocessor chips. Initially Ti and TiN barrier layers are deposited, followed by
chemical vapor deposition of W to fill the contact vias. Proceeding from achieving local
and global planarization of SiO2, removal of excessive tungsten from the horizontal
surfaces on the wafer pattern proved to be an asset for subsequent Al metallization.
Hence CMP was developed with a two-fold approach of (a) planarizing oxide and (b)
removing the via fill metal from the horizontal surfaces. Along with its successful
implementation for the achievement of the above-mentioned objectives, CMP is now
extended to (a) polishing of different metals like Al, Cu, Pt, Au, Ti, Ta, etc., (b) polishing
of different insulators like SiO2, Si3N4, various low-k dielectrics, doped and undoped
oxides of silicon, (c) polysilicon, (d) ceramics like SiC, TiN, TaN, etc., (e) multichip
modules, (f) packaging; (g) optoelectronic components, (h) flat panel displays, (i)
microelectromechanical systems (MEMS), and (j) magnetic recording heads and CD read
write drives [19]. The major applications of CMP are summarized in Table 1-2.
Cmp Output Parameters
The critical output parameters of CMP are described below. Depending on the
application, different output parameters are optimized.
Table 1-2. Chemical mechanical polishing applications.
Materials Application
Metal
Al Interconnection
Cu Interconnection
Ta Diffusion barrier/adhesion promoter
Ti Diffusion barrier/adhesion promoter
TiN, TiNxCy Diffusion barrier/adhesion promoter
W Interconnection e-emitter
Dielectric
Cu-alloys Interconnection
Al-alloys Interconnection
Polysilicon Gate/interconnect
SiO2 ILD
BPSG ILD
PSG ILD
Polymers ILD
Si3N4 or SiOxNy Passivation layer, hard mask
Other
Aerogels ILD
ITO Flat panel
High k dielectrics Packaging
High To superconductors Interconnection/packaging
Optoelectronic materials Optoelectronics
Plastics, ceramics Packaging
Silicon on Insulator (SOI) Advanced device/circuits
Ref [19]
Removal rate
The polishing steps involving CMP of STI and metal mainly emphasize removing
an overburden of one material and stop on another material leaving a planar surface. A
high material removal rate is one of the key outputs desired from such CMP processes.
The removal rate of the material being polished is strongly influenced by slurry chemistry
and pressure distribution, i.e., chemical action of the slurry chemicals on the material,
interplay of the different completing agents, oxidizers and corrosion inhibitors and the
mechanical abrasion of the particles on the wafer. Chemicals such as oxidizers and
corrosion inhibitors vastly affect the reaction rate of the slurry with similar particle
nature, size and distributions.
1 0 5% H202, lOn1 M 0n
S8 BTA %
DI Water
2-
0 10mM BTA
0.00 0.01 0.03 0.05 0.07
Time (sec)
Fig. 1-5. Surface layer formation on Cu with various slurry chemistries. [20]
Fig. 1-5 shows the variation of the reaction rate of the different slurry components
on Cu when the reaction kinetics were studied using electrochemical chronoamperometry
[20]. It can be seen from the figure that the surface rate kinetics reach approximately 60
A/s when Cu is immersed in distilled water. The reaction rate increases to around 120 A/s
when 5% H202 is added. However upon addition of 10 mM BTA, the reaction rate
decreased considerably. Therefore, the chemical contribution is very important in the
CMP process. The other important factor is the down pressure used. However, for
effective CMP of materials, the enhancement of removal rate by increasing the down
force should not be at the cost of surface quality (for example, increase in wafer to wafer
and within wafer non-uniformity, delamination, dishing, erosion, etc.) [21, 22] of the
wafer. For this reason, the process throughput is compromised to polish the wafers at
lower applied load, thereby achieving global planarization and better surface quality.
Global planarization
In contrast to glass or silicon wafer polishing, the objective of the CMP process in
interlayer dielectric (ILD) polishing is to planarize the topography created by previous
semiconductor processing steps. The pad and slurry play key roles in achieving global
planarization, concurrently maintaining the removal rate and producing a defect free
wafer surface.
If mechanical removal predominates, high frictional forces are produced which
result in surface defectivity. Also variation in local polishing pressure leads to variable
removal rates within the wafer, which seriously compromises global planarization [20],
whereas, if chemical removal dominates, etching and corrosion [23] can occur. The key
to a good polishing process is dependent on synergy between chemical etching and
mechanical planarization. For this purpose, a passivation layer is created on the wafer
surface by reaction with slurry and removal of this passivation (or modified) layer occurs
due to mechanical abrasion of slurry particles (Fig. 1-6). For example, in Cu polishing,
the formation of the passivation layer is achieved by oxidizers such as hydrogen
peroxide, potassium ferricynate, ferric chloride, and ferric iodate. Corrosion damage to
the surface is prevented by adding corrosion inhibitors such as benzotriazole (BTA). In
tungsten polishing, a rapid formation of surface passivation layer is due to the use of
peroxygen compounds and stabilizing agents [24]. The purpose of the passivation layer in
the case of silica polishing is to soften the surface, which is inherently hard resulting in
minimal material removal. For this purpose, maintenance of alkaline pH is sufficient in
most cases [25, 26]. During Ta polishing, formation of stable Ta205 helps in uniform
removal of material from the surface [27]. To achieve optimal CMP performance, the
time to form the passivation layer of desired thickness should be minimized.
Abrasive partsic
Fig. 1-6. Micro and nanoscale phenomena during CMP [20]
Selectivity
The difference in the removal rate of one material versus another in a given system
represents the measure of the slurry selectivity for these two materials. The higher the
slurry selectivity, the more effective the end point detection is for the given CMP process.
This is due to a marked change between the tribological properties of the material being
polished, and the underlayer. The difference in tribological properties can be monitored
in-situ using techniques such as motor current, force or acoustic emission for better end
point detection. Selectivity is a very important criterion in designing any slurry for
multilevel metallization. For STI, the slurry must affect the oxide that is being planarized,
and not the underlying nitride layer. For polishing Cu, the slurry needs to react selectively
on Cu and spare the barrier layer Ta and underlying layers of silica or low-k dielectric
material. Typically slurry selectivity of 10-25 has been reported for Cu polishing [20].
The selectivity could be considerably improved up to 1000 by introducing a particle free
slurry [28, 29]. The reduction in mechanical component due to lack of abrasive implies
that the majority of material removal occurs due to the solution chemistry which can be
made highly selective.
Addition of specialized chemicals that can act as catalysts in the chemical
interaction between slurry and the polished material thereby increasing the rate of the
chemical reaction considerably is a common approach for improving the slurry
selectivity. Tetramethyl ammonium hydrate can be added to Cu slurries to considerably
decrease the silica polishing rate [30]. An increase in the polishing rate of TaN with
addition of phosphoric acid in alumina and colloidal silica slurry is also observed. Mixed
abrasive slurries (MAS) containing alumina/silica and alumina/ceria particles have been
developed to improve the selectivity of the CMP slurry. The details of the experiments
with these novel mixed abrasive slurries have been published by Jindal et al. [31, 32].
Though highly selective slurries might help in avoiding defects such as erosion of sub
layers, planar defects such as dishing still exist causing wafer-to-wafer nonuniformity
(WTWNU) due to the differential polishing pressures in the process.
Measurement Techniques
The outputs of the CMP process are quantified by a host of metrics. Material
removal rate (e.g., nm/min) is commonly measured in terms of thickness loss per time or
weight loss per time. In terms of global planarization at the wafer scale, film thickness is
measured at points across the wafer surface and used to calculate within-wafer-non-
uniformity (WIWNU). Most commonly, a 49-point measurement grid is used to
determine the WIWNU, which is the variation of those measurements with respect to the
average thickness. Similarly, a within-die non-uniformity (WIDNU) calculates a similar
variation but just across the area of one die. At the feature-scale of the process, metrics
such as step height reduction are used to indicate efficiency of the CMP process.
Selectivity of the slurry is more often characterized by estimating the erosion or polishing
of the underneath layer.
End Point Detection
Theoretically, the end of a polishing step can be realized by setting a time limit to
the process. However, this is only a rough estimate because variations of polishing pad
surface property, fluctuations in slurry chemistry, particle size distribution, feeding rate,
pad conditioning issues, and other potential variables can cause over- or under-polishing.
Therefore, in-line monitoring or automatic end-point detection of CMP is needed. Initial
oxide CMP relied on process times and control of removal rate stability to achieve target
thicknesses. As the CMP environment is hostile for in-situ metrology, the earlier endpoint
techniques relied on indirect methods. One of the commercialized indirect methods
monitors the motor current change [33] while driving the carrier or platen rotation. This
technique measures changes in friction forces between the pad and the wafer surface. It is
mostly successful for tungsten CMP than for copper or STI CMP processes, because the
frictional forces between the polishing pad and exposed films do not differ significantly
in the case of copper or STI scenarios. The method sufficiently sensitive for STI,
dielectric and metal CMP processes is the vibration monitoring method [34]. Vibrations
during the CMP process are detected using an accelerometer on the carrier behind the
wafer. This method is used to detect delamination of low-k dielectric films in copper dual
damascene process [35].
Optical reflectometry is the preferred method for endpoint detection. But, the
method is highly intrusive, requiring extensive polisher modifications. Various other
endpoint techniques have been developed in recent years, many of which are currently
adapted to the high volume production stage [35].
CMP Consumables
The CMP market in the year 2003 was $1.7B with slurry (24%) and pad (15%)
amounting to 40%. The CMP equipment was estimated to comprise 55% with the
remaining 6% accounting for conditioner and other consumables. The market is expected
to grow to $3.3B in the year 2008. The rapid growth of CMP and its application for
polishing various materials, necessitated development of a variety of slurries, pads,
complex process recipes, more complex slurry mixing and distribution systems. Polishing
of different materials and customized needs of the various semiconductor industries
resulted in more complex CMP equipment with different process dynamics such as
linear, orbital and fixed head machines [35]. The process of CMP has gone through a lot
of evolution from first being used just for silicon dioxide planarization to the present day
planarization applications in pre-metal dielectric (PMD), ILD, STI, metal and gate oxide,
etc.
While the CMP process seems fundamentally easy to understand, there are a
staggering number of process variables (Fig. 1-7) that complicate the underlying
interaction between the pad, slurry and wafer, preventing full comprehension of the
process. The most widely accepted mechanism for the material removal is that the
chemicals in the slurry react with the surface of the wafer (e.g., oxide, copper, etc.) and
create a "modified" layer that can be removed easily. The mechanical action of the
abrasives trapped in pad asperities then physically abrade and remove the modified
material [24, 36, 37].
In terms of the CMP input parameters, the process variables are generally
categorized into chemical and mechanical parameters. The chemical nature of the CMP
process typically relates to the constituent makeup of the slurry and how those chemicals
react with the wafer surface to create a passivated or modified layer, keeping the particles
well suspended in the slurry. The chemicals in the slurry are highly dependent upon the
type of CMP being used. For example, in oxide polishing, an alkaline solution is used
along with various buffering and oxidizing agents, whereas for metal CMP as in Cu
CMP, acidic slurry can be used. The mechanical aspects of CMP largely depend on the
polishing pad, wafer pattern, and process conditions like pressure and relative pad-wafer
velocity.
Slury
" Slurry chemistry
Oddizers
Compi exing agents
pH stabilizers
SPar licl e features
Size distribution
Mechanical strength
Slurry stabilizers
Shape
Slurry agglomeration
*Slurry theology
Wafer
* Wafex material properties
* Feature characteristics
Pad
* Padroughness
* Mechanical properties
* Conditioning technique
Tool
* Load distribution
* Platen rotation
* Wafer rotation
* Slurry flow rate
Fig. 1-7. Parameters involved in CMP process.
C
Polishing in its simplest sense is controlled chemo-mechanical material removal to
produce a globally flat, defect free surface. The main components of CMP are the CMP
slurry and the pad. The complexity of these CMP components and their interactions are
discussed further in the following sections.
CMP Slurry
CMP slurry consists of two major components: abrasives and chemicals.
Depending on the material of the wafer, the chemistry of the slurry, and the synergy
between them, each kind of slurry behaves differently [38]. CMP is influenced to a great
extent by numerous slurry parameters such as pH, solution chemistry, particle charge
type, concentration and particle size distribution of abrasives, completing agents,
oxidizers, buffering agents, surfactants, corrosion inhibitors, etc.[19, 39]. An ideal CMP
slurry should be well suspended with good slurry stability, and be able to achieve high
removal rate, excellent global planarization, prevent corrosion (in case of metal,
especially Cu), and result in good surface finish, low defectivity and high selectivity.
Rate of surface reaction
The reaction between the wafer and slurry chemicals to form a modified layer of
desired thickness should be fast enough to avoid material removal by mechanical
indentation of the bare wafer. The time scale in which the passivation layer is formed
should be less than the average time of successive particle indentations into the wafer for
abrasion to achieve optimal CMP performance. Fig. 1-8 shows the electrochemical
chronoamperometry (potentiostatic) analysis of Tungsten by first keeping the samples at
cathodic potential to avoid surface oxidation and then anodizingg" them. The generation
of current, which corresponds to surface reaction rate, is monitored on a millisecond scale
as shown. It can be seen form the figure that Tungsten surface quickly passivates which is
conducive for mechanical removal of the material.
r 10,000
+L.7V
1.v
I A
10 -0.2
0.Q0[ 0,001 0,01 0.1 1
Time (s)
Fig. 1-8. Chronoamperometry measurements for tungsten [20]
For selectivity purposes, the reaction of slurry chemicals with the underneath layer
should be significantly low compared to the surface layer. In terms of operational hazard,
the chemicals used in the slurry should not make it hostile for the operators performing
CMP and the post-CMP chemicals should be treated to avoid environmental damage.
Slurry agglomeration
The ideal slurry will have abrasives behaving as discrete single particles during the
CMP process. However, particles in a real CMP slurry apart from being discrete can also
exist in the form of agglomerates [38, 40]. An agglomerate is a collection of particles that
come together into packed clumps due to physical or chemical interactions and act as a
single entity. Although CMP slurries are composed of very fine particles up to 200 nm
(0.2 [tm), "large" particles of 1-3 |jm and greater are often observed during the process.
Such particles can be due to agglomeration or contamination from foreign material.
These are not desirable for CMP as they can cause micro scratches and pits due to deep
indentation or non-uniformity due to varying polishing pressure. Even minimal and
undetectable concentration of larger sized particles are shown to produce surface defects
in CMP [41]. The agglomeration of slurry can be prevented by proper slurry design,
which includes appropriate use of chemicals, and/or surfactants and the suitable order of
their addition to design the slurry [42]. Other than adding chemicals, the particle surfaces
can also be altered to make the suspension stable during the CMP process [42, 43].
In order to prevent the particle agglomeration during the distribution stage,
dispense filters are used which filter out the agglomerated particles before slurry delivery.
Care is taken so that the filtration of the agglomerated particles does not change the
particle distribution and slurry composition. The drop in slurry pressure and flow due to
filtration is minimized by using a series of filters instead of a single filter.
Chemical Mechanical Polishing Pads
The pad traps the slurry particles on its asperities, and executes the polishing action
by transmitting the normal and shear forces for polishing, thereby playing a very crucial
role in process optimization [19]. The polymeric polishing pad must have sufficient
mechanical integrity and chemical resistance to be effective for numerous polishing runs
through its life. The pad must be able to withstand the aggressive slurry chemistries used
in CMP without degrading, delaminating, warping or blistering. Slurry chemistries range
from highly alkaline near pH 11 used for polishing of inter-layer dielectric to highly
acidic near pH 2 used for metal polishing. Slurries also contain oxidizing agents such as
hydrogen peroxide and potassium iodate. The pad material should be resistant to such
environments. As the polishing slurries are aqueous based, pad should be hydrophilic to
allow circulation of slurry to whole wafer during polishing.
The polishing pad is made up of a matrix of cast polyurethane foam with filler
material to tailor the hardness of the polyurethane impregnated felts for the specific CMP
process. Polyurethanes have a unique property of combining high strength to resist
tearing during polishing and, an acceptable level of hardness and good abrasion
resistance to prevent excessive pad wear during polishing. The CMP pads can be
categorized into four types differentiated by their microstructure [44]. They are: Polymer
impregnated felts (type 1), Poromerics (type 2), Filled polymer sheets (type 3), and
Unfilled textured polymer sheets (type 4). Pads of type 1 and 2 are shown in Figs. 1-9
and 1-10. The figures indicate the variety of pad surface textures or asperities involved in
CMP. The polyurethane polymer consists of two major components/blocks of alternating
hard organic isocyanates and soft polyols. Each of these segments has different
properties, the characteristic properties of a given pad depend on the relative ratios of
these hard and soft segments. Isocyanates, the more rigid blocks, influence the hardness,
modulus, tear strength and stability, specifically at elevated temperatures. Polyols on the
other hand tend to affect the elasticity, toughness, abrasion resistance and the low
temperature stability of polymer.
For uniform polishing, the pad needs to be in contact with the whole wafer surface
with the same contact pressure and relative velocity between pad and wafer in all contact
regions. But wafers are not flat and typically have some degree of curvature due to
stresses induced during manufacture process or differing coefficients of thermal
expansion of the various deposited oxide and metal layers on the substrate. This
necessitates the polishing pad to be flexible enough to conform to the wafer-scale
flatness. This is achieved by laminating a stiff polishing pad to a flexible underlying base
pad, which is typically a more compressible material. This improves polishing uniformity
across the wafer without unduly compromising the stiffness of the top polishing pad. A
stacked pad consisting of an IC 1000 top layer on a Suba IV sub-pad (Rodel Inc.,
Newark, DE) is currently the pad system of choice to obtain both good global wafer
uniformity and high removal rate for various CMP systems.
(a) (b)
Fig. 1-9. Type I CMP pad. (a) Surface. (b) Cross-section [44].
(a) (b)
Fig. 1-10. Type II CMP pad. (a) Surface. (b) Cross-section [44].
Effect of pad grooves
Polishing pads have macrostructure, which can be either perforations through the
pad thickness or surface groove designs. The motivation behind designing pad grooves is
to prevent hydroplaning of the wafer, ensure uniform distribution of slurry over the
wafer, and to remove the polishing debris from the pad surface through these chemicals
[44]. The groove designs can be circular with either concentric or spiral grooves, cross-
hatched patterns arranged as an X-Y grid across the pad surface, other regular design
such as hexagons, triangles and tire-tread type patterns, or irregular designs such as
fractal patterns, or their combinations. The effect of pad texture on tribological and
kinetic properties of polishing pad has been studied by Philipossian and Olsen [45]. Real
time monitoring of coefficient of friction (COF) was done to estimate the normal shear
forces originating during a particular CMP processes for a given pad. Frictional and
removal rate data were taken on Rodel IC 1000 flat, perforated, XY- and K-grooved
pads. The Somerfeld number, COF, a new parameter called tribological mechanism
indicator (TMI) and materials removal rate were some of the parameters monitored
during the experiments. Philipossian and Olsen [45] reported that flat and XY-grooved
pads begin in boundary lubrication and migrate to partial lubrication as Sommerfeld
numbers increase. Thus different pad surface textures showed different material removal
mechanisms. However, the variation of the mechanical properties need to be studied
further on a microscale to gage its impact on the CMP process and material removal
mechanism.
Physical and chemical changes in pad during polishing
Since polishing is a wet process, the mechanical properties of the pad must not vary
considerably when the pad is wet during CMP. However, the hydrophilic pad material
looses modulus and hardness and gains ductility because water acts as a plasticizer. These
changes are reversible [44] and pad properties revert to the dry values, when the pad is in
dry state, indicating that the chemical attack is minimal [46]. The surface of the
polyurethane pad has been studied and reported in literature [47]. The quantitative
analysis of the physical and chemical changes that occur in the polishing pad during
polishing was performed by Lu et al. [48]. In their research, Lu et al. studied the effect of
polishing cycles on the pore size and shape of the commercially available IC 1000/Suba
IV polishing pad. It was reported that the pore size and shape distribution were modified
by polishing cycles. Polishing induced a permanent distortion in the radial direction due
to the oscillatory motion of the wafer, while no change was seen in pad pore size in the
transverse direction (direction of pad motion). The change was attributed to pore closure
by surface reflow of the polymer during polishing or conditioning and not due to debris
filling. The surface profiles generated using white light interferomery showed that the
overall micro roughness of the pad decreases with polishing and there is a smoothening
effect over a period of time due to polishing cycles. Along with physical changes, the pad
undergoes surface chemical modification and degradation due to polishing slurry
chemicals. No pad degradation of the bulk material of the polishing pad was observed.
The change in spectrum (Fig. 1-11) and shifting of the peaks of the used pad as compared
to the new one has been attributed to the realignment of the molecules of the damaged
pad surface layer.
The porous structure of the polishing pad encourages water seepage. The presence
of water in the polymer pore significantly affects the mechanical properties of the
polishing pad. Li et al. [47] studied the effect of pad soaking time on the mechanical
properties of the polishing pad using dynamic mechanical analysis (DMA). The dynamic
shear modulus decreased to two thirds of its original value when the pad is soaked for 5
h. Further, the rate of decrease in dynamic shear modulus decreases at around 14 h but
does not reach a steady state. No significant change in the removal rate was observed
when pads with different soaking time are used in oxide polishing. This shows that
removal rate is more of a surface characteristic of the pad and is not affected by the bulk
material.
slurry. [48]
Modification of the pad-wafer contact mechanics can directly impact the CMP
process performance. The modified surface can even change the mechanism of material
removal during a polishing process [49]. During polishing, the pad surface layer degrades
decreasing the pad surface roughness. The pad surface which is flattened or "glazed"
needs to be rejuvenated to bring the polishing performance back to the original value.
131)1.1) 12511 12(11 I 1 ) IlflO I [| ) I{H) W, I
Fig. 1-11 Troughening absorbancf the spectrad by a diamof used nditioner is known to result in pad revival.
slurry. [48]
Modification of the pad-wafer contact mechanics can directly impact the CMP
process performance. The modified surface can even change the mechanism of material
removal during a polishing process [49]. During polishing, the pad surface layer degrades
decreasing the pad surface roughness. The pad surface which is flattened or "glazed"
needs to be rejuvenated to bring the polishing performance back to the original value.
The roughening of the pad by a diamond conditioner is known to result in pad revival.
Along with wafer polishing, the conditioner made of diamond grit considerably alters the
surface of the polishing pad. The process of conditioning essentially removes some
amount of top pad surface and is performed either in-situ (during the polishing run) or ex-
situ in between two polishing runs. The pad damage due to insufficient conditioning is
termed as "glazing". Experiments were performed on the CMP pad by Lawing [49] at
different conditioning levels of aggressiveness (expressed as low, medium and high). The
change in polishing rate correlated better with the predicted surface contact area rather
than the average surface roughness.
Effect of temperature change on polishing pad
The polishing pad might be subjected to elevated temperatures during the course of
the CMP process [19]. The heating is due to the frictional forces generated by solid-solid
contact between pad and wafer during polishing. There is some alleviation of heating
when the polishing is done in hydrodynamic contact mode [48]. However, there is an
overall increase in the pad temperature irrespective of the mechanism of polishing. The
effects of pad heating are compounded if the chemical reaction between slurry and pad is
exothermic [38]. The mechanical, physical and chemical properties of the polyurethane
material can be permanently or temporarily altered if the pad is heated beyond a
particular temperature [50]. DMA, thermomechanical analysis (TMA) and
thermogravimeteric analysis have been used for material characterization of the polishing
pad [47, 48].
The applicability for DMA for polishing pads has been elaborately studied by Lu et
al. [48]. Based on the results of previous experiments performed by Li et al., [47], DMA
was performed to evaluate the structure-property relationship of polyurethane pads. The
glass transition temperature is the temperature at which the macromolecules of the
polyurethane break away from entanglements and undergo micro-Brownian motion.
Investigation of glass transition temperature of the polyurethane pad is important as it
corresponds to a sharp decrease in the hardness and modulus of the pad which in turn
affects the removal rate and global planarization.
Non-uniformity of the polishing pad
The pad might be directly responsible for several process defects like WTWNU
where there is non-homogeneity of polishing when one wafer is compared to another or
WIWNU where there is non-homogeneity of polishing at different areas of the same
wafer. The scanning ultrasound transmission (UST) is a nondestructive technique
developed that works on the principle of ultrasound permeability through absorbing
visco-elastic medium [51, 52]. The difference in the ultrasound absorption in the areas of
varying density and viscoelasticity is used to determine the non-uniformity within a
single pad there by giving an in-depth idea of the physical characteristics of the given
pad. In order to improve the yield of the CMP process and, to get a highly planar, defect
free uniform wafer surface with reduced overall manufacturing costs involved, there is a
need to extensively study the fundamental properties of the CMP pads on the whole.
CMP process in the future
The semiconductor industry has effectively adapted its CMP technology for the 300
mm wafer [53]. Beyond the adoption of copper interconnects, several technologies are
necessary to continue the shrinkage of device dimensions and the increase of packing
density in ULSI manufacturing. The use of ultra low-k materials as interlayer dielectrics
has been at the forefront for decreasing the "C" part of the "RC delay". However,
polishing of ultra low-k dielectrics which are mechanically weak (soft) is a daunting task.
The single and dual damascene structures comprising of ultra low-k porous or polymeric
materials are more prone to buckling and crushing failures due to significantly lower
hardness and Young's modulus as compared to silicon dioxide. According to the ITRS
roadmap [12] materials with dielectric constant 2.2 will be integrated in the IC by year
2007.
Also, there needs to be a marked improvement in slurry selectivity for accurate end
point detection when the constituent layers of the damascene structure namely, metal,
hard mask, cap layer, barrier layer and dielectric are polished. Furthermore, it is
necessary to explore the niche of the CMP process in STI and other applications such as
backside polishing and the fabrication of micro-electro-mechanical systems (MEMS).
Here, the CMP process must be integrated horizontally and vertically to achieve high
throughput and performance. In MEMS applications, horizontal integration ensures
reliability and good performance of a specific CMP process run. The development of new
polishing pads and pad architectures, novel slurries, new metrology techniques, etc.
comes within the scope of horizontal CMP integration. Vertical integration ensures
success of every successive CMP operation. This includes the integration of the upstream
processes such as Cu/barrier deposition, etching and downstream processes such as ILD
deposition and lithography. This opens a wide scope of research for further optimization
and development in Cu CMP process. Due to intricate interactions of the processes
involved in any CMP process, modeling efforts are needed to get a systematic
understanding of the process and predictability of the CMP outputs.
Scope of Dissertation
The importance of polishing and the various available planarization techniques
with emphasis on CMP have been discussed in Chapter 1. A general overview of the
parameters involved in CMP are discussed with prominent interactions between them.
Chapter 2 gives an overview of the CMP modeling efforts and the framework of the
PERC I CMP model. The model's challenges are described, which necessitated the
29
development of the PERC II model. The assumptions of the PERC I model are refined as
necessary in chapter 3. Model development is also delineated with critical differences
between PERC I and II models. Chapter 4 validates the PERC II model comparing to
experimental literature data and comments on the predictive capability of the model.
Chapter 5 summarizes the findings of the study along with the suggestions for future
work.
CHAPTER 2
CMP MODELING: OVERVIEW OF PERC I
Like most semiconductor manufacturing processes, simultaneous tuning of
process variables is a formidable task due to many varying parameters. Unfortunately,
much of the commercial CMP process development cycle is based on nearly blind
design-of-experiment (DOE) testing. While these empirical tests help to optimize
parameters in a very small range, the overall approach does not necessarily facilitate a
deeper understanding of the process. Thus, CMP research has focused on developing
process models in order to gain further insight on the process mechanics for better control
over the CMP outputs.
The ultimate goal is to model at three scales of CMP, namely, the particle, die and
wafer scales. The models at the particle scale are needed to understand the mechanisms
involved in interactions between slurry particles, chemicals, polishing pad and wafer
during blanket wafer polishing. The models at the feature or die scale are needed to
address the topography evolution of integrated circuit (IC) chips as a function of pattern
density, line width, pitch width and polishing time. The wafer-scale models also need to
address the issues related to the material removal non-uniformity over the entire wafer
surface. To satisfy the manufacturability requirements in the sub-micron IC fabrication,
the margin of the profile non-uniformity is usually limited in the range of several hundred
nanometers. The material removal non-uniformity across the wafer is rooted in the non-
uniform distributions of pressure and velocity over the wafer surface, which in a large
degree, are determined by the configurations of CMP equipment. A model at this scale,
therefore, needs to consider the equipment configurations appropriately.
Literature Review
In the last decade, various kinds of models have been proposed by researchers,
concentrating on different aspects of CMP covering the particle, die and wafer scales. But
the effectiveness of die and wafer scale models depends strongly on the robustness of the
particle scale models, as the die and wafer-scale models are developed based on the
particle scale models. Therefore, in order to understand and predict the performance of
CMP in wafer scale, the current study aims to understand the behavior of CMP at particle
scale. As the current work focuses mainly on particle scale modeling, only related models
will be reviewed here.
Preston Equation
The Preston equation [54] as an empirical relation for optical glass polishing was
substantiated by Cook [24] who considered that the interaction between abrasive particles
and wafer surface leads to removal of atomic clusters due to bond breakage. Preston
provided a simple model of material removal in glass polishing tools which is often
adopted for CMP processes also. But, experimental results have shown that slurry and
pad properties also significantly influence the process along with down pressure and
platen speed [55, 56]. In order to explain the non-Prestonian behavior, many modeling
efforts have been carried out in the past decade and is still a topic of considerable interest.
The three most important elements in the CMP material removal process are the
wafer being polished, slurry and polishing pad. Researchers adopted two main
approaches in modeling the CMP process. One approach is based on fluid
hydrodynamics, where material removal is considered to be caused mainly by slurry
erosion. Slurry film thickness, hydrodynamic pressure and removal rate are generally
calculated by solving Reynolds or Navier-Stokes equation [57-62]. The other approach
uses the theory of contact mechanics, which assumes that the material removal occurs by
particle indentation into the wafer surface. Pad surface roughness and hardness are
generally used to calculate the particle indentation into the wafer thereby calculating the
removal rate [27, 63-67]. Finite element modeling (FEM) has also been used to determine
the pad asperity mechanical properties recently [68, 69].
It is generally agreed that the typical CMP process conditions lead to major
material removal by particle abrasion rather than by slurry erosion [70]. Experimental
observations of polished surfaces have revealed that material removal in CMP occurs as a
consequence of a combination of chemical reaction of the slurry chemicals with the wafer
surface and ploughing (repeated sliding, rolling, or indentation) of the abrasive particles
against the wafer surface that is already modified. The chemical reaction results in
formation of a thin layer on surface which is of different modulus than the
substrate/wafer, aiding in material removal [71].
When particles are introduced between wearing surfaces, the wear is due to the
particles embedded in one of the sliding surfaces. This process is referred to as three-
body abrasion. In the CMP process, the particles acting as single point cutting tools, dig
into the wafer and plow a series of grooves in it. Therefore, the CMP process is
essentially a three-body abrasive wear. The analysis of the texture of polished surfaces
also supports this observation. Fig. 2-1 illustrates typical post CMP surfaces as examined
via an optical surface profiler. The surface texture consists of a number of randomly
oriented nano-grooves (Fig. 2-1). The width and depth of the grooves are determined by
the penetration depth of the abrasive particles into the wafer surface.
301
' 75,00
S2,5
f-
5.0 nm
2.5 nrm
0.0 nm
111111111-- -.. -
0 2,50 5.t
C1m
(b)
Fig. 2-1. Polished wafer surfaces showing nanoscratch surfaces. (a) Silica wafer [72]. (b)
Si wafer [73].
The material removal process in CMP can be considered as sliding of the indented
slurry particles over a chemically influenced thin layer over the wafer surface. The
knowledge of critical factors that effect the CMP process and the interactions among
them is critical to select the respective parameters and design an enhanced CMP process,
predicting material removal rate, and surface quality (roughness, defects, etc.). The most
popular material removal equation is the Preston's Equation [54], which was initially
introduced for glass polishing,
MRR= KePV, (2 1)
where MRR is the material removal rate, Ke an all-purpose coefficient, P the down
pressure and V the relative velocity over the wafer-pad inteface. It demonstrates a linear
dependency of material removal rate on the pressure and velocity. Not all experimental
MRR data in CMP, especially in metal CMP, support the linear dependency between
material removal and the product of pressure and velocity.
Modifications to Preston Equation
In order to explain the non-Prestonian behavior of the CMP process, revised
Preston's equations were proposed by several researchers. For example, considering that
the material removal rate does not extrapolate to zero, Maury et al. [74] introduced a
fitting parameter MRRo into Preston's equation
MRR= KePV + MRRo (2.2)
Later, the nonlinear equation
MRR= KePu VO, (2.3)
where ac and P are two fitting parameters, was proposed by Wrschka et al. [75] to get a
better fit for the experimental data.
The major problem with Preston's equation and its revisions are that they are
empirical modeling efforts. Therefore, they do not shed any light on gaining insight into
the process especially regarding other process variables. By extending Brown's model
[76] of the metal polishing to the silicon polishing, Cook [24] developed a physical
model to address this limitation. The interactions between the abrasive particles and the
wafer surface is proposed as a Hertzian elastic penetration [77] of a spherical particle
under uniform pressure P into the wafer surface, sliding along the surface with a velocity
V and removing material volume proportional to the penetration. The MRR formulation
was proposed was
MRR= PV/(2E) (2.4)
where E is the Young's modulus of the wafer material. This model can be taken as a
theoretical equivalent to the Preston's equation since it supports the linear dependency of
MRR on pressure and velocity. The relationship between the wafer surface roughness Ra
the down pressure P, and abrasive size can also be obtained based on this model
Ra=3/4 b (P/2klE)2/3 (2 5)
where kl is the particle concentration, which is unity for a filled close hexagonal packing
[24] and b the diameter of the slurry particles. A similar model was developed by Liu et
al [78] based on the statistical method and Hertzian elastic penetration. Besides the wafer
material properties including wafer hardness Hw and wafer Young's modulus Ew, this
model included pad hardness Hp and abrasive Young's modulus Es resulting in
MRR = C, H,- E, + E PV (2.6)
H +H, P Es Ew
where Ce is a coefficient to account for the effects of slurry chemicals and other
consumable parameters. This model, similar to Cook's model, suggests that the material
removal is proportional to the applied pressure and relative speed
The advantages of Cook's and Liu's models over Preston's equation are that they
provide more insights into the roles of the consumable parameters. The contributions of
the slurry abrasives and pad, for example, have been attributed to their size and hardness.
An additional benefit is that not only material removal rate, but also surface quality issues
such as roughness (Eq. 2.5) can be addressed using these models.
Slurry Erosion
In Cook's and Liu's models, the mechanical removal by abrasive particles was
assumed to be the dominant mechanism. However, some researchers instead assumed the
material removal to be by "slurry erosion" approach, i.e., due to mechanical-enhanced
erosion. Runnels et al. [79-81] developed an erosion-based model for CMP. They
assumed that a fluid film exists between the wafer and pad interface, which affects the
erosion/material removal rates at each single point through the fluid stress tensors.
Motivated by observations at feature scale, modification to Preston's equation was
proposed, with relative velocity V, replaced by the tangential stress on the wafer surface;
MRR=Ceoton (2.7)
where Ce is an all purpose coefficient, ot is the shear stress due to the slurry flow and on
the normal stress.
The model uses a steady state incompressible Navier-Stokes equation to model the
slurry flow at the wafer-polishing pad interface. The wafer surface is assumed to be
smooth and spherical with a large radius of curvature. The fluid layer thickness and the
angle of attack between the pad and the wafer are obtained through an iterative procedure
satisfying force and momentum balance. Runnels' model has been integrated into several
particle-scale models by researchers including Tseng and Wang [82] and Zhang and
Busnaina [83]. Tseng and Wang attributed the normal stress at the particle-wafer contact
to the elastic indentation of the particle into the wafer surface, which is similar to that
proposed by Cook and calculated the normal stress over the wafer-particle interface as
F
n (2.8)
where F is the force acting on the spherical particles, which is proportional to the pressure
P. The radius of particle-wafer contact, re is given by
r3 3d 1 2 1 v '2
r 8 F+ Er- (2.9)
8 E E'
where d is the diameter of particles, v and v' the Poisson's ratios of wafer surface and the
particle and E and E' the elastic modulii of the wafer and particles, respectively. The
shear stress due to the slurry flow can be approximated as
t =CeIVPA, (2.10)
where |t is the dynamic viscosity of the slurry and Ao the area of wafer surface
respectively. Substitution of equations (2 8) and (2 10) into (2 7) yields
MRR=KeP5/6V1/2 (211)
where Ke is the parameter to account for material properties, slurry abrasive
concentration and chemical processes. This model demonstrated a nonlinear relationship
between the material removal and the product of pressure and velocity. Tseng's model
attempted to connect the elastic indentation to the erosion rate, whereas Cook's and Liu's
models connect it to the mechanical abrasion. While the down pressure dependency (P5/6)
is quite close to linear dependency, the velocity dependency (V2) however, is quite
nonlinear. This is because the contribution of velocity has been attributed to the slurry
flow instead of abrasive sliding. Due to pad porosity and surface asperities, such a slurry
film is bound to be interrupted by solid-solid contact. Thus hydrodynamic lubrication
may be responsible for distributing the slurry, while the material removal occurs mainly
through solid-solid contact.
Cook's, Liu's and Tseng and Wang's models attributed the penetration of the
abrasive particles to Hertzian elastic contact. Zhang and Busnaina [83] estimated the
contact pressure between the particle and the contact surface in their model and found
that it is larger than the yield stress of the polished materials. Therefore, they proposed
that plastic deformation is a more likely deformation mechanism in polishing. The
contact pressure over the particle-wafer interfaces is suggested to be equal to the hardness
Hw of the wafer materials. Replacing the normal stress (Eq 2 8) in Tseng and Wang's
model with the hardness Hw yields the following material removal rate formulation
MRR= Ke(PV)1/2 (2.12)
where Ke is the parameter to account for materials properties, slurry abrasive
concentration and chemical processes. Both Tseng and Wang; and Zhang and Busnama's
models suggest a nonlinear pressure times velocity dependency of material removal rate.
Moreover, it is noted that besides the external force applied on the particles from the pad,
Zhang and Busnaina also proposed that an adhesion force, either van der Waals force or
electrostatic force depending on the separation distance between the particle and the
wafer, contributes to the indentation. This has been integrated into another particle-scale
model by Zhang et al. [84]. In a series of articles by Ahmadi and Xia [85], and Mazaheri
and Ahmadi [86, 87], a thermodynamic work parameter Wa of adhesion was added to
account for its effects on the indentation of abrasive particles. Mazaheri and Ahmadi [86]
introduced a double layer (dl) force into the indentation force, whose magnitude is a
function of the zeta potential of the abrasives and the wafer. Beside the abrasive wear,
Ahmadi and Xia [85] also considered the adhesion wear of the wafer in their model.
Moreover, while most of the models treated the abrasives as having a spherical shape,
Mazaheri and Ahmadi represented them as spheres with a number of hemispherical
bumps around their surface. The penetrations of slurry abrasives are modeled using the
penetrations of the bumps.
The above models imply that the abrasives are embedded in the pad and indent into
the wafer surface. It should be noted that some modeling efforts also assume that the
abrasive particles float in the slurry and impact the wafer surface from time to time
causing material removal [88].
Pad Considerations in CMP Models
The pad is assumed to be smooth in the earlier particle-wafer interaction models. It
has been observed that the pad topography and pad material play an important role in
material removal processes. For example, the material removal rate has been reported to
increase with the pad surface roughness [55]. Without conditioning of the polishing pad
surface, the material removal rate decreases exponentially with polishing time [89]. In
consideration of this, Yu et al. [63] developed a pad-based model. They approximated the
peaks on the pad surface by hemispherical asperities with constant radius and the asperity
heights to follow Gaussian distribution. Based on the model, the real contact area is
smaller than the nominal (apparent) contact area and linearly proportional to the down
pressure. Steigerwald et al. [19] proposed that the material removal rate is proportional to
the number of abrasive particles over the contact area. Combining this argument with
Yu's model yields a linear dependency of the material removal rate on the down pressure,
which agrees with Preston's equation.
Zhao and Shi [90] also proposed a model based on the wafer-asperity contact.
Unlike Yu's model [63], this model does not consider the Gaussian distribution of the
asperity heights. The contact area between an asperity and the wafer is given by Aa Ua P2/3
based on Hertz elastic contact theory. By combining Steigerwald's argument, the material
removal rate formulation can be obtained as
MRR= K(V)P(2/3) (2.13)
where K(V) is a function of the relative velocity V and other CMP parameters. It is
further considered by Zhao and Shi [90] that when the particles are rolling against the
wafer surface, their contribution to material removal will be negligible. They argued that
whether the particle is rolling or not is determined by the surface friction between the
particles and the wafer, and only when the down pressure P is larger than a threshold
down pressure Pth, the pure rolling can be avoided. The material removal rate is given as:
MRR = K(V)(P2/3 Pth2/3) for P > Pth (2.14a)
and
MRR = 0 for P< Pth (2.14b)
The fundamental difference between the above pad-based models and the particle-
based models by Cook and others is that pad-based models attribute the material removal
rate to the number of abrasive particles captured by the polishing pad while the later
attributes the material removal rate to the interaction between a single abrasive and the
wafer.
A particle-pad interaction is critical to the particle scale models considering that the
function of the pad is to trap the abrasive particles and transmit load to the particle-wafer
interface. Cook's model [24]suggested a close-packing of spherical abrasives onto the
pad. It is assumed that the wafer and pad are separated completely by the abrasives and
no direct pad-wafer contact exists. The force applied on a single abrasive under these
assumptions is given by
F = Pr2 (2 15)
ki
where P is the polishing pressure, r the abrasive size and kl the particle fill fraction on the
pad. This particle-pad interaction model has been integrated into the material removal
model of Cook. It is also used by Ahmadi and Xia [85] to evaluate the force on a particle
in their case of a hard-pad and a larger concentration of abrasive particles. Later, Zhao
and Shi [90] proposed that when the pad is soft enough, the abrasive particles will be
embedded deeply into the pad and the force from the wafer is supported by the pad and
abrasives together. This idea has been applied by Luo and Domfeld [27, 91, 92] and Fu et
al. [93] in their model. Luo and Dornfeld's model [27, 91, 92] suggested that this force is
proportional to the contact pressure multiplied by the abrasive size by assuming that the
abrasives are closely packed to each other and these closely packed abrasives are
enclosed by the pads so that the effective contact area between wafer and pad is equal to
that without abrasives. Moreover, the size of the abrasives that may be captured by the
pad is assumed to be a function of abrasive size distribution and pad properties. Their
model, however, neglects the lower sizes of the particle size distribution in material
removal calculation. Fu et al. [93] later assumed that the abrasives are dispersed evenly
over the pad surface and used a beam model to evaluate the wafer-pad direct contact
between two single abrasives. The force supported by a single abrasive can be obtained
from the beam model and is a function of abrasive size, down pressure and pad material
properties [93].
Besides the force, the second purpose of an abrasive-pad interaction model is to
evaluate the number of abrasives involved in the material removal process. Fu et al. [93]
considered the number of trapped particles as an independent parameter in their model.
This may be misleading considering that various parameters, for example, the abrasive
weight concentration and abrasive size, have an influence on the number [27, 72, 91, 92].
Luo and Dornleld considered more complex scenarios and suggested that only a portion
of abrasives are involved in the material removal process, which is not reasonable.
Slurry Considerations in CMP Models
The contribution of slurry chemicals to the material removal is either neglected in
earlier models or represented by an all-purpose coefficient. Cook [24] suggested a
complete but complicated scenario of the chemical effects. Besides the mechanical
removal, it is proposed that the surface removal during polishing include the following
five chemical processes (i) the slurry chemical diffusion into the wafer surface, (ii) the
subsequent wafer material dissolution under the load imposed by the abrasive particles,
(iii) the adsorption of the dissolution product onto the surface of the polishing grain, (iv)
the re-deposition of the polishing materials back onto the wafer surface, and (v) the
corrosion rate between particle impacts. These steps are not considered in most of the
early models as they are complex to model quantitatively.
The major contribution of slurry chemicals on the material removal process has
been attributed to the formation of a surface layer. This idea is well demonstrated by
Kaufman et al. in their Tungsten CMP model [37].
Kaufman et al. proposed that this passivation layer is removed by slurry abrasives
and the fresh tungsten surface is exposed, which is subsequently passivated and removed.
This mechanism of continued passivation and removal can be used to explain the copper,
aluminum and other metal CMP systems as well. A similar mechanism is supposed to
work in silicon, silica and low-k CMP systems. Paul [94] and Zhao et al. [95] have
proposed detailed surface kinetics models to connect the slurry chemical concentrations
and fresh metal surface sites available to the formation rate of the surface layer. Their
models can explain the material removal rate as a function of chemical and abrasive
weight concentration. Borst et al. [62] have proposed a five-step model for CMP of SiLK
dielectrics (i) mass transport of reactant from the slurry to the slurry/wafer interface, (ii)
adsorption of reactant to available surface site, (iii) reaction between adsorbed reactant
and specific wafer surface to form an altered wafer surface layers, (iv). mechanical
removal of the altered wafer surface layer, and (v) mass transport of removed material to
the bulk slurry. In their work, the mass transportation, slurry chemical concentration and
reaction rate to the formation of the surface layer were incorporated. This detailed model
was suggested to be able to be extended to the CMP of other materials.
The idea of continuous surface modification and removal has been applied in the
model developed by Qin and co-workers [96]. A similar approach has also been used in
Luo and Dornfeld's model [91]. They extended Kaufman's model by proposing that the
surface layer is a bi-layer structure, though in reality there might be a gradual transition
between the layers. They did not cover the details on how the formation rate of the
surface layer is affected by the slurry chemicals. But Luo's model assumes that all the
asperities contact the wafer and that the particles fill the asperity-wafer contact spots,
which is not necessarily true.
The slurry chemicals affect not only the wafer but also the slurry abrasives and
polishing pad. Their contribution to material removal is therefore not only reflected
through surface kinetics, but also through alteration of the abrasive and pad properties
such as the abrasive shape, abrasive size and pad Young's modulus. There are not many
modeling efforts on these two interactions yet. A recent model by Castillo-Mejia et al.
[97] tried to explain the effect of water-pad interaction in CMP. The water is proposed to
plasticize the polishing pad and reduce its elastic modulus. A formulation on the ratio of
the Young's modulus of wet pad to the dry pad is suggested. This Young's modulus is
then used in a wafer-pad contact model to evaluate the material removal.
Many of these models were also extended to patterned wafers along with other
modeling concepts including combination of the solid-solid contact and slurry erosion
approaches [98-109]. Polishing mechanisms were also based on surface reaction kinetics
[95, 110, 111] and modeled using molecular dynamics simulations [112-115].
PERC I CMP Model
Most of the earlier models have concentrated on one or two interactions of the
CMP process parameters. They are useful for identifying the input parameters. However,
they may not be sufficient for understanding the whole material removal process. The
effects of the same input parameters may be contrary to each other when acting in
different conditions. A comprehensive model of CMP integrating the main interactions
together is therefore needed. For this purpose, a model was developed by Qin et al.[72],
taking into consideration most critical CMP parameters. This section discusses the
assumptions on which the model is built, its effectiveness in predicting the CMP process
variables and the model limitations.
Model Development
The chemicals in a slurry alter the properties of the wafer surface that is in contact
with them. Since the chemical influence is restricted to a thin region near the wafer
surface, a thin surface layer whose mechanical properties are different from those of the
bulk wafer is assumed. This surface layer can be formed by chemical reactions at the
surface or by the diffusion of slurry ionic species into the wafer. The surface layer is then
removed by mechanical action of the slurry particles and/or by dissolution into the slurry
through a convective diffusion process. Although there may be a few exceptional cases
[Luo98], material removal by the dissolution is small compared to that by the mechanical
action of the slurry particles, and it is not considered in this model. Fig. 2-2 describes
schematically the physical situation that is considered in the present analysis. The pad
asperities that are in contact with the wafer surface act as the anchoring sites for the
slurry particles. The anchored particles then gouge the chemically altered wafer surface
or the bulk wafer depending on the indentation depth resulting in material removal. In the
following section, various parameters considered in the present analysis and the
underlying assumptions are described followed by the details of the model development.
Topology of a pad surface has been characterized and described statistically by
various researchers [63, 116] Their measurements indicated that pad asperity heights
were in the order of tens of microns and appeared to have a normal (or Gaussian)
distribution. In Fig. 2-2, the topology of the pad surface is shown schematically along
with a statistical description characterizing the pad asperities. In this figure, 3 is the
mean radius of curvature of the tip of the pad asperities, 4 is the distribution function for
the height of the asperities, and cy the standard deviation of the pad asperity heights,
respectively.
Asperity Height, z
(b)
Fig. 2-2 Pad consideration in PERC I. (a) Schematic of the pad surface that is in contact
with the chemically modified layer formed on the wafer surface; (b)
Distribution of the pad asperity heights (When the height of an asperity is
greater than the average gap h between the pad and the wafer, that asperity is
in contact with the wafer surface.) [72]
Since the characteristic length scale for the roughness of the pad surface (e.g., 3) is
much larger than the feature sizes on the wafer, it was assumed that the wafer surface is
smooth and flat compared to the pad surface. This assumption makes it possible to use
the Greenwood-Williamson model [117] that describes the contact area, Areal, between a
flat surface and a randomly rough surface under a normal force (or load), L:
A ,real = TNA z h) (z)dz (2.16)
L = NAE12(z h)3/2 ()dz (2.17)
where N is the number density of the pad asperities (i.e., number of asperities per unit
area), Ao is the total area of the flat wafer surface, and h is the average gap between the
smooth wafer and the rough pad under the given conditions. Then, the down-pressure, P,
applied to the wafer with an area Ao is L/Ao whereas the actual mean contact pressure, Pr,
is L/Areal if the normal force applied to the wafer is entirely supported by the pad
asperities in contact with the wafer. That is
L =PA,= PrAreal (2.18)
E* is the composite modulus between the two surfaces in contact, that is defined as
1 1-v2 _V2
1 ~ (2.19)
E* Ep E
where E and v are the Young's modulus and the Poisson's ratio, and the subscripts p and
w represent the pad and the wafer, respectively.
Equations 2.16, 2.17 and 2.18 provide an explicit relationship between the contact
area (Ar) and the down pressure (P) that is needed in modeling the CMP process. From
Eqs. 2.16 and 2.17,
AL 4E* 3 h h)/2(z Jz (2.20)
L 4E (z -h)3/( Z
Following Yu et al.[63], the normal distribution function may be used for the
distribution of the asperity height, ). When )(z) = exp -z-- is substituted into
Cy [27j 2CY2 2
Equation 2.20, the integral cannot be obtained in an explicit form. Instead, a numerical
integration should be preformed. Johnson showed that the ratio of the two integrals in Eq.
2.20 is nearly a constant for a broad range of h/o [77]. Thus,
rea =-C 1 =C-1, PA, 22
A-C C (2.21)
where C is a constant whose value is between 0.3 and 0.4 when h/l is in the range of 0.5
and 3.0. Recalling that o is the standard deviation of the asperity height distribution,
h/o=3.0 represents a situation where only 0.13% of the pad asperities are in contact with
the wafer whereas about 30.85% is in contact when h/o=0.5. Practical situation in a
typical CMP process is well within this range ofh/l value. From Eqs. 2.18 and 2.21,
Pr = C E (2.22)
It should be noted that the contact area (Ar) increases linearly with the down-
pressure whereas the contact pressure (Pr) remains constant due to the linear increase of
Ar with P.
The underlying assumption for Eq. 2.18 is that the applied normal force on the
wafer is supported entirely by the pad asperities. As is mentioned previously,
experimental measurement on the friction coefficient indicates that the hydrodynamic
pressure is negligible under typical CMP operational conditions. Furthermore,
commercial pads often have grooves or perforations to prevent hydroplaning (i.e., to
prevent a large pressure development due to a lubrication flow) and to ensure uniform
slurry distribution. Thus, the assumption that the hydrodynamic pressure is negligible is
justifiable.
Small Plastic Deformation of the Thin Layer on the Wafer Surface
Considering the fact that the deformation of the surface layer by the slurry particles
under the contact pressure is very small in the order of several nanometers, the mode of
deformation can be assumed to be plastic deformation. In the case of oxide CMP, it is
known that the silica surface is softened when it is in contact with an aqueous solution. A
surface layer (or so-called gel layer) is formed due to the diffusion of water molecules
into the silica surface altering the surface properties.
The situation may be different in the case of metal CMP. Since an oxidizer is added
in most metal CMP processes, metal oxide is formed on the wafer surface creating a
structure mismatch at the interface between the oxide layer and the bulk wafer
underneath. Thus, delamination or microcracking of the oxide layer may occur at the
interface during the metal CMP process. However, as evidenced in previous sections, it is
likely that plastic deformation is still the dominant mode of deformation because the
indentation depth by the slurry particles is small (e.g., less than several nanometers).
Furthermore, the oxide layer may be soft due to hydration in an aqueous environment.
Thus, the assumption that the chemically altered thin surface layer on the wafer exhibits
plastic deformation is plausible. This assumption enables us to apply the abrasive wear
theory in deriving the expression for the material removal rate. When the indentation
depth is beyond the surface layer, the deformation of the unaltered wafer material may be
either plastic or elastic depending on the hardness of the wafer material and the slurry
particles.
Material Removal by a Single Particle
In the present analysis, it is assumed that the slurry particles are monosized and of
radius 'R'. The overall material removal rate is assumed to be the multiple of the removal
rate by a single particle and the number of particles in contact with the wafer:
(RR) = n (RR)v/Ao (2.23)
where n is the number of slurry particles anchored in the region of the pad that is in
contact with the wafer, and (RR)v is the volumetric removal rate by a single particle. The
overall removal rate, (RR), is typically described in terms of the thickness removed per
unit time; hence the division by the nominal wafer area Ao.
Schematic illustration of single particle abrasion is given in Fig. 2-3. The material
removal is viewed as a sliding-indentation process in which the volumetric removal rate
by a particle is given as
(RR)v = AiV (2.24)
where Ai is the cross-sectional area of the particle immersed in the wafer under pressure
(hashed area in Fig. 2-3a), and V is the relative velocity of the pad to the wafer.
Assuming that the hard particle maintains its spherical shape,
A, = -a(R o)+ R2 arcsin a
R (2.25)
where co is the indentation depth and a the radius of the circular indentation on the wafer
surface by the particle of radius R,
Because co is very small compared to R, a/R is also very small and Equation 2.25
can be linearized about a/R to give
A, = ac +
3 I R]) (2.26)
Thus the volumetric removal rate is
(RR), = ao V
3 (2.27)
WaferWafer
t
a Co t '"
b Pad
PPad t
tFa P Fa
(a) (b)
Fig. 2-3. Schematic of single particle abrasion showing the removal of chemically
modified surface layer. (a). o < t. (b). o > t with soft wafer materials.
Because a2=(2Rco-o2) and co is much smaller than R, Equation 2.27 can be
rewritten as
2a3V
(RR) 2aV
3R (2.28)
Once the indentation depth co (or a) is known, the volumetric removal rate by a
single particle can be readily calculated from Eq. 2.28.
Depending on the relative magnitude of the indentation depth (co) to the surface
layer thickness (t), the material removal may be classified into two different regimes as
depicted in Fig. 2-3; (a) o < t and (b) co > t.
Indentation depth smaller than the surface layer thickness (i.e., o _< t)
When the chemically modified surface layer is thick enough, the indentation depth
co will be no greater than the surface layer thickness and only plastic deformation within
the surface layer occurs. In this case, o) (or a) can be obtained from the force balance
imposed on the particle by the pad and the wafer that depends on the indentation depth of
the particle into the wafer surface. The force imposed on a particle by the pad is given as
F = 7rb' (2.29)
where b is the radius of the circular indentation into the pad as described in Fig. 2-3a, and
P, is the contact pressure given by Eq. 2.22. The force on the particle by the deforming
wafer can be given as
F = Ta 2B (2.30)
where Be is the effective Brinell hardness of the chemically modified surface layer. The
term "effective" is used because its value is influenced by the thickness of the chemically
modified surface layer and the material property of the underlying substrate. The pad is
supposed to hold the particles in place thereby transmitting the normal force to the
particles that cause indentation of the wafer surface. The pad should be soft enough to
provide stable anchoring of the particles on its surface. However, if it is too soft, the
particles will be embedded too deeply into the pad diminishing the material removal
capability of the particles. Thus, the pad should be sufficiently hard to make the value of
b in Fig. 2-3 to be close to the particle radius R. Then, from equations 2.29 and 2.30 with
b R,
1/2
a=R (2.31)
This equation along with equations 2.22 and 2.28 gives the following expression for the
volumetric removal rate by a single particle of radius R.
(2 "-3/4 E 3/2
(RR) C3/2R2 ~/2 V (2.32)
S3 P Be
Indentation depth greater than the surface layer thickness (i.e., o > t)
When the bulk wafer material beyond the surface layer is relatively soft (e.g.,
copper) compared to the hardness of the particles, it is possible for the particle to cause
deformation into the wafer beyond the surface layer (Fig. 2-3b). As it was described in
the previous section, the deformation is expected to be plastic for both the surface layer
and the bulk wafer, and the force causing the deformation may be given as
F, = [2tRB, + (a2 2tR)B j (2.33)
where Bw and Be are the Brinell hardness of the bulk wafer material and the chemically
modified surface layer, respectively. Because this force is equivalent to the one imposed
on the particle by the down-pressure, Eq. 2.33 is equated with Eq. 2.29 with b=R to give
R2P, + 2(B, Bj)Rt 12
a R + 2(B (2.34)
Then, Eq. 2.28 for the volumetric removal rate by a single particle is reduced to
(RR) =2C(j R2 +21 Rt (2.35)
3 B BR
While equation 2.35 describes a situation where the deformation reaches into the
bulk wafer beyond the surface layer, deformation of the bulk wafer may not occur if the
bulk wafer material is very hard. This limiting case corresponds to the situation of
infinitely large Bw in equation 2.35 that is reduced to
(RR) =4, R1/2t3/2 (2.36)
3
Equations 2.32, 2.35 and 2.36 indicate that the volumetric removal rate by a single slurry
particle depends on the geometry of the pad and the particle (i.e., 3, C and R), the
material properties (E*, Bw and Be), the thickness of the surface layer (t), and the relative
velocity of the pad (thus the particle) to the wafer (V). It may be interesting to note that
(RR), is independent of the down pressure, P. It is due to the fact that the contact
pressure, Pr, is independent of P. As equation 2.21 indicates, the contact area of the pad
with the wafer (Ar) increases linearly with P. Consequently, the mean contact pressure,
P, which is PAo/Ar, is independent of P. In the following section, the number of particles
in contact with the wafer surface is estimated to determine the overall removal rate.
Estimation of the Overall Removal Rate
When the wafer is pressed against the pad and is in sliding motion relative to the pad, the
particles can be entrapped only in the region where the local distance between the wafer
and the pad asperity is smaller than the particle diameter, 2R. Thus, the number of
particles that can be trapped on the pad asperities is
n = 2R'Areal" Pn
(2.37)
Here A, is the real contact area between the wafer surface and the pad asperities as
it was described in previous section. p, is the number density of the particles in the slurry
(i.e., number of particles per unit slurry volume), that can be calculated from the solid
loading Ca (wt%) of the slurry particles as
3 Cap,
P, (2.38)
4 7R3 pa
where pa and ps are the densities of the particle and the slurry solution, respectively.
Substitute of Eqs. 2.38 into 2.37 gives
3 Cap9
n = A, (2.39)
2 7R2pa
By combining Eqs. 2.39 and 2.18, the number of particles anchored on the pad asperities
becomes
3 Ca s 0 pAA
n- 2C -CR (2.40)
2C 7R2 pa a E
Now that (RR), and n are given in terms of various processing variables and material
properties, the overall material removal rate in a CMP process can be estimated from Eqs.
2.32, 5.27 and 2.40 for the two different cases described above:
Removal Rate: Indentation depth smaller than surface layer thickness (i.e., co
(RR= C P (E)2Be32 PV (2.41)
Removal Rate: Indentation depth larger than surface layer thickness (i.e., co > t),
(RR) = -C1 {c U+ 21 1-1 3/2 PV (2.42)
v zC pA, El ) B B) R
As Eq. 2.23 indicates, the overall removal rate, (RR), is the thickness removal rate which
is the volumetric removal rate divided by the nominal wafer area Ao. We may recall that
C is a constant whose value is about 0.35 when h/o is between 0.5 and 3.0 [Joh85]. The
overall removal rate given above is reminiscent of Preston's equation in that the material
removal rate is proportional to the down-pressure, P, and the relative velocity, V. It
should be noted that the Preston's coefficient Kp, whose value relies on empiricism, is
now given as a function of various processing variables and material properties.
The model predicts the removal rate as a function of pressure and velocity product (PV)
very well [72]. The removal rate is usually only considered as a function of PV in many
modeling efforts, ignoring other processing parameters. But the PERC I model considers
the particle size effect, which is crucial for any CMP system.
Equation 2.41 indicates that the overall removal rate is independent of the particle size R
whereas Eq. 2.42 indicates a decrease in (RR) with increasing particle size. When the
indentation-depth is smaller than the surface layer thickness (i.e., 0)
removal rate by a single particle (RR), is proportional to R2 (Eq. 2.32). The number of
particles in contact, on the other hand, is inversely proportional to R2 as the Eq. 2.40
indicates. Consequently, the particle size effect is negated.
Thus the model predicts as shown in Fig. 2-4 that the removal rate is either
independent of particle size in the modified layer removal regime or decreases with
particle size in the substrate indentation regime. But, the reported literature has shown
57
that removal rate also increases with particle size in a given range, and that there is an
optimal particle size at which enhanced polishing can take place as shown in Figs. 2-5
and 2-6. To overcome these limitations, a (PERC II) model is developed by analyzing the
crucial assumptions in PERC I, and refining them when necessary. The following chapter
deals with assumption validation and PERC II model development.
Modified
layer
removal
Wafer
removal
Partide Size
Fig. 2-4. Schematic ofPERC I predictions for increasing particle size.
0 10 20 30 40 50
Mean Particle Size (nm)
60 70
Fig. 2-5. Polishing of silica by nanoslurry showing optimal particle size [118].
200
150
E
* 100
o
E 50
a-
600
-*- 18 kPa, 0.4 m/s
5500 -- 32 kPa, 0.6 m/s
-0 45 kPa, 0.9 m/s
400
i 300
B~200
100
0
25 60 95 130
Mean Particle Size (nm)
Fig. 2-6. Polishing of silica at various pressure, velocity and particle sizes showing
optimal particle size [116].
CHAPTER 3
PERC II CMP MODEL
The previous chapter highlighted the basic concepts involved in PERC I model
[96] development, along with various challenges faced by it. In this chapter, some of the
critical model assumptions are verified and refined as necessary. The changes are
incorporated into the basic framework of the PERC I model to develop the PERC II CMP
model.
Assumption Validation
The critical components in the CMP process are the wafer, pad and slurry. As the
wafer is fixed in a given CMP process, the contributions from pad and slurry are
investigated to include their most crucial characteristics into the model. The main role of
the pad in CMP is to trap the abrasive particles for material removal and transmit the load
applied onto the trapped particles. In particle scale modeling, the function of the pad in
global planarization need not be considered.
Pad-wafer Contact Area
The pad-wafer contact area strongly influences the number of particles that can be
trapped under each pad asperity. The greater the contact area between pad and wafer, the
higher the number of particles that can be trapped resulting in enhanced material removal.
The changes in pad properties during polishing are important to model the CMP process.
Often dry properties of the pad are investigated [119], rather than the more appropriate
wet properties. The pad characterization investigations are also limited to the bulk [48,
68, 120] of the material rather than the surface. These bulk properties in dry state are
usually employed in conventional CMP modeling efforts [68]. Since, polishing involves
the interfacial behavior of pad, abrasive, and wafer, pad surface properties should be used
rather than the bulk properties. Surface properties of the pad rapidly change due to
immersion in slurry [121, 122], quickly reaching equilibrium values [46].
There is one fundamental question that needs to be answered. What is the real
contact area between pad and wafer? And is the Greenwood-Williamson model [117],
which states that "contact area is linearly proportional to the applied load" applicable in
CMP systems? A basic assumption of the PERC I model is that removal rate is linearly
dependant on the pad-wafer contact area. Abrasive particles trapped between pad
asperities and wafer are considered to be responsible for material removal. As applied
load on the wafer increases, the pressure at the pad asperity-wafer interface is supposed
to remain constant due to elasticity of the pad, while the pad-wafer contact area increases.
This behavior is predicted by the "Greenwood-Williamson" model, which assumes that
the pad asperity tips contact the wafer under Hertz elastic theory conditions. Although a
number of CMP models have employed this concept [63, 109], no systematic attempts
are reported that provide validity of this assumption. Similarly, some researchers have
assumed the contact area to be proportional to the power of two-thirds [90] of pressure
without any experimental basis. In the past, pad-wafer contact area has been generally
derived mathematically [27, 123]based on statistical surface roughness rather than actual
experimental measurements. The pad-wafer contact area reported in literature including
theoretical calculations varies from 0.0005% to 20% [27, 49, 63, 109, 124, 125].
Considering the importance of reliable value of the pad-wafer contact area and its
dependence on applied load in CMP modeling, a systematic attempt was made to
measure it under CMP conditions. Kim et al. [124] have measured the pad-wafer contact
area using optical microscopy and have reported a value of 5%. On the other hand, Basim
[125] has reported the value to be 0.3% in dry state based on FTIR measurements. To
investigate the pad-wafer contact area, a reliable technique is needed. To establish the
reliability of optical microscopy and/or FTIR techniques, contact area measurements
were performed on pad using optical microscopy.
Optical microscopy
The average pad asperity radius is around 30 |jm for a typical CMP pad and the
RMS roughness around 5[im. Therefore, an optical microscope can be used to study the
contact spots, the majority of which can be expected to be over a micron in size. To
simulate the wafer, a microscope coverslip was used. To minimize the bending of the
coverslip under applied load, a fixture was fabricated as shown in Fig. 3-1 which enabled
easy access to the microscope objective lens for translation in "in-plane" direction. The
fixture consists of a base plate with a window hole of 10mm x 5mm to allow optical
observation of the contact spots with minimum bending of the coverslip under applied
load.
Optical measurements were performed in dry environment using Leika microscope
equipped with frame grabber. One inch square coupons were punched out from IC 1000
pad, which was used for polishing. The pad was ex-situ conditioned using a grid-abrade
diamond pad conditioner. Minimal contrast between contact and non-contact regions was
obtained, when the pad specimens were analyzed under the given load. This was not even
sufficient for qualitative analysis. Measurable distinction between contact and non-
contact spots was achieved with coating of the pad surface with a 5-10 nm thick gold-
palladium layer using the Denton sputter coater. The thin coated layer is not expected to
affect the pad-wafer contact area results, as the attempt here is to measure the contact
area under dry conditions.
Cover sl
Metallic Base
Microscope objective
Fig. 3-1. Schematic of the fixture used in measuring pad-wafer (dry) contact area.
Measurements were performed in two modes. Mode I consisted of measuring the
contact area of selected asperities as a function of applied load and mode II consisted of
measuring various contact spots under applied load by translating the microscope
objective to cover the entire specimen in view. The optical micrographs were analyzed
using Spot and Image-Pro software to calculate the percent contact area.
The micrographs obtained in mode I are shown in Fig. 3-2. The regions "P" and
"N" indicate the pad asperities that are of interest in all three figures, though only Fig.3-
2a is labeled for clarity. In "P" regions, the asperity contact increases with applied load,
whereas, in "N" regions, new asperity contacts are formed with applied load. The pad-
wafer contact area values obtained from optical measurements are plotted in Fig. 3-3. The
results indicate that the contact area of a given pad asperity increases linearly with
applied load and more importantly the magnitude of contact area obtained from optical
technique is in the same range as obtained from the FTIR technique by Basim (Fig. 3-4).
The discrepancy between this data and Kim's optical microscopy result [124] could be
due to the possibility that some non-contact areas were counted as contact spots due to
poor contrast, as no pad surface modification or coating was mentioned in their study.
1--I
10 micron
(a) (b) (c)
Fig. 3-2. Pad asperity contact spots seen through optical microscope. (a) 0.75 psi. (b) 5
psi. (c) 8.8 psi.
A Various asperity contacts at same pressure
S0.8
S0.6 -i
0.2
0
S0.4 A
0.2 0 .
0 1 2 3 4 5 6 7 8
Applied pressure (psi)
Fig. 3-3. Pad asperity contact area dependence on applied load. Each legend represents a
selected asperity contact.
0.55 :--* ,-- --- ,-- ---,- i "-'--1-- ---, ----,
0.55 .- I I
0.45
0.40 -
0.35 a0.33 ~, pad contact
^ 0.35
o 03" u
-e0.1505 .., .
0.15 *
0.10
0 2 4 6 7 8 10 12
Head Pressuie (psi)
Fig. 3-4. Pad contact area as a function of applied load [125].
The main motivation behind investigating contact area measurements is to get a
realistic value of pad-wafer contact area under CMP conditions. The optical and FTIR
techniques are used in dry conditions, whereas the actual CMP is a wet system.
Combining optical data from this study and FTIR data from Basim [125], it is established
that FTIR with attenuated total reflection (FTIR-ATR) method can be successfully used
to estimate the pad-wafer contact area. The same technique was employed in the present
investigation to measure the contact area in wet condition simulating the CMP
environment.
Fourier transform infrared spectroscopy (FTIR)
To investigate the pad-wafer contact area variation with applied load in various
slurries, FTIR-ATR was used. The contact spots contribute toward the absorbance signal
resulting in quantitative estimation of the pad-wafer contact area depending on signal
intensity. Maximum signal intensity is obtained when the pad surface is in full contact
with the wafer surface. The contact area is obtained by normalizing the measured signal
intensity with the maximum signal intensity [126]. The setup used for contact area
measurements is shown in Fig. 3-5.
Applied load
IR Beam
ATR crystal
Fig. 3-5 Schematic of the FTIR-ATR technique used to measure pad-wafer contact area.
A silicon ATR crystal is used in contrast to the ZnSe crystal employed by Basim
[125] to simulate the silica wafer system. A nitrogen purged Nicolet Magna 760
spectrometer equipped with liquid nitrogen chilled MCTA detector was used to carry out
FTIR analysis for better accuracy. All the spectra were the result 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 silicon crystal. The intensities of the CH2 peaks of the
polyurethane were recorded in the wavenumber range of 3000-2750 cm 1. The
asymmetric and symmetric stretches of CH2 at wavenumbers of 2927 and 2855 cm-1
respectively were analyzed. The contact area is expected not to depend on pad-wafer
relative velocity, hence they are held stationary. The slurries were simulated by using
solutions of varying pH with no other chemicals (passivating agents) or particles added.
This is valid especially in silica CMP as no passivating chemicals are generally added in
slurry formulation. The absence of particles reduces the errors due to scattering. The
water FTIR/ATR signal is subtracted from pad spectrum (both at same pH) to get the
desired signature spectrum (Fig. 3-6). A representative piece of pad sample (IC-1000)
soaked in given solution for one hour was placed on the (Si) ATR crystal and loaded at 4,
7 and 10 psi in the wet state under solution. The solutions studied are nanopure water
adjusted to pH 4, pH 9 and pH10.5 with either 0.5M HNO3 or 0.5M KOH.
Silica CMP is generally performed at slurry pH 10.5, as the silica solubility
increases drastically after pH 10. Thus, there should be minimum silica wafer
modification or "modified layer" formation at pH 4 and pH 9. Whereas, pHl0.5 should
lead to considerable surface dissolution of silica. These conditions are chosen, as all CMP
process variables except pad properties are supposed to be constant in pH range of 4 to 9,
and increased surface modification (dissolution) at pH10.5. The typical absorbance
spectra obtained for a given specimen at different applied loads is shown in Fig. 3-7. The
signal intensities at symmetric and asymmetric stretches are averaged to calculate the
contact area. The results obtained for different pH and dry conditions indicate that contact
area increases linearly with applied pressure for all conditions (Fig. 3-8). The normalized
plot shows the accuracy associated with the measurement technique, which is consistent
for all the conditions used. The linear dependence of pad-wafer contact area on applied
load validates the use of Greenwood-Williamson theory in the CMP model. Pad-wafer
contact area in the wet state was measured to be around 2-3 times the contact area in the
dry state (Fig. 3-9). This implies that the removal rate calculated would be 2-3 times less
if the dry contact area is used in the model. There is no discernible change in pad-wafer
contact area under different pH conditions due to data scatter, possibly due to uncertainty
67
associated with subtraction of water spectrum. Utilizing the value of pad-wafer contact
area obtained and the Greenwood-Williamson assumption, pad asperity modulus is
derived which is used in calculation of pad asperity stress distribution.
SWater: pHl0.5
J V,,
Signature peak
Fig. 3-6. Analysis of the FTIR-ATR spectra.
lopsi
e' 10psi
I I psi
90
2927. 12855
Wavenumber (cm-1)
Fig. 3-7. Absorbance spectra of pad (in pH 4) at different applied pressures.
Pad: pH10.5,7psi
ViL^
2.5
2
1.5
1
0.5
5 7 9
Pressure (psi)
Fig. 3-8. Contact area variation with applied load in dry and wet conditions.
1.2
1
0.8 -
0.6 -
0.4 -
0.2 -
0-
I Pressure = 10 psi
Dry
pH 4.0
pH 9.0
pH 10.5
Fig. 3-9. Contact area in dry and wet systems in various pH conditions.
--Dry
---- pH 4
- -pH 10.5
Atomic force microscopy (AFM)
Contact area increase in the wet state is attributed to changes in the mechanical
properties of the pad. Consequently, surface mechanical properties of pad were
investigated using a colloidal probe atomic force microscope (AFM), where a particle is
attached to AFM tip to measure pad properties. A Digital Instruments Dimension 3100
AFM equipped with a liquid cell adapter was used.
Particle indentation into pad was used to assess pad surface mechanical properties
at pH levels of 4, 9 and 10.5. It was realized that a stiff cantilever is needed to probe the
surface mechanical properties of the pad. Hence a tapping mode cantilever was modified
by cutting the tip area with focused ion beam (FIB) yielding a cantilever with a stiffness
of 75 N/m as compared to 30 N/m before the modification. A 7[tm silica particle was
attached to the tip of cantilever for measuring mechanical properties of the pad.
Nanoindentation measurements were carried out using the modified AFM tip shown in
Fig. 3-10. The pad surface was scanned with the colloid probe to find a relatively flat
surface before indentation data was measured. This is expected to minimize the effect of
surface asperities or roughness on the surface mechanical analysis. Typical indentation
data are shown in Fig. 3-11. Only the approaching curve is used to investigate the pad
surface mechanical properties. The results from AFM are summarized in Fig. 3-12.
The indentation data presented in Fig. 3-12 indicates that the pad surface
mechanical response is markedly different in dry and wet systems. In dry pad case, the
force on AFM probe increases with indentation depth at a constant pace without any
change in stiffness. Whereas, for wet cases at all pH levels employed, a dramatic change
in slope is observed. The indentation data is not reproducible over entire pad surface, as
the pad surface is not perfectly smooth due to asperities. But all the indentation data in
wet state showed the stiffness change in contrast to the dry data. When the indentation
data is aligned such that all curves match the dry pad behavior, it can be seen that the pad
surface is affected by the solution pH which results in different mechanical response
compared to the bulk material (Fig. 3-13). The moduli of pad surface and bulk shown in
Fig. 3-12 were estimated utilizing Hertz theory. The modulus of dry pad is atleast 3-5
times the pad modulus in dry state indicating pad swelling.
(a)
Silica parali
(b)
Fig. 3-10. Colloidal probe AFM tip (a) Modified tapping mode cantilever, dotted line
shows the region of the cantilever removed by FIB. (b) Silica particle attached
to the modified cantilever.
70
--Approaching
50 + Retracting
e 30 Air i Pad
O 0
0
U-
10
-10 f--
-200 0 200 400
Indentation Depth (nm)
Fig. 3-11. Force-distance curve indicating softer "modified layer" on the pad surface.
The surface modification of polyurethane is not entirely unexpected. The
polyurethane polishing pads possess both -COC- and -COOC- functional groups in the
main chain structure which are susceptible to hydrolytic attack [50], especially in alkaline
media [50]. Disruption of hydrogen bonding is reported for pad samples immersed in DI
water for minutes [97]. Schematic of hydrogen bond disruption is shown in Fig. 3-14.
This hydrolytic attack can explain the pad surface mechanical property behavior.
The presence of surface modification on pad surface or pad swelling could impact
the abrasive trapping ability of the pad asperities along with the load transmitted to the
particles during polishing depending on particle size. The change in contact area
observed in wet state indicates that depending on slurry chemistry, the pad-wafer contact
area could be effected, directly influencing the material rate. Hence, when different slurry
chemistries are used in a CMP process, the CMP model needs to incorporate the slurry
effect not only as a modified layer on wafer surface, but also as change in pad-wafer
contact area and abrasive trappability, which has not been accounted for in past CMP
modeling efforts.
100
80
0 50 100
150
Normalized Indetation Depth (nm)
Fig. 3-12. Force-indentation curves from AFM for different conditions. All curves are
offset such that they match in the top portions to clearly show the evidence of
a softer layer on pad in wet systems.
Bulk pad
Softer layer
Bulk pad
(b)
Fig. 3-13. Schematic of pad surface as deduced from AFM data. (a) Dry state. (b) Wet
state.
H H
Hydroged -
Bodg V H
II 1I
0 0
(a) (b)
Fig. 3-14. Schematic of polishing pads (a) Structure showing hydrogen bonding between
neighboring polymer molecules, (b) Water can form hydrogen bonds with
polymer molecule by disrupting hydrogen bonds between the polyurethane
molecules [127].
Pad Asperity Height Distribution
A quantitative understanding of the pad-wafer and pad-particle interface is
fundamental to a comprehensive CMP model. The pad asperities are assumed to remain
constant through the planarization of each wafer. It is known that the asperities involved
in polishing wear out changing in height over time, while the shorter asperities, those that
do not come into contact with the wafer, retain their original height. Thus the pad
roughness or the asperity height distribution changes with polishing, if pad conditioning
is not performed properly. The surface roughness of pad is known to be directly
proportional to the polishing rate [128-130]. Different types of pads have different
asperity height distributions, though they might have similar average surface roughness.
In PERC I, pad asperites were assumed to be spherical and of same size with pad asperity
height averaged out to obtain the pad asperity stress, and pad-wafer contact area. This
implies that the pad asperities are not only assumed to be equal in size and shape, but also
of equal height. This imposes the pad to have contact spots equal in size, experiencing
same contact pressure. Thus, all the particles trapped in these contact spots have same
indentation into the wafer surface leading to equivalent material removal under each
asperity. Thus the previous PERC model was equivalent to having a single asperity
model with no effect of asperity height distribution. Depending on the CMP system and
desired planarization criteria, different pads are used in practice. Therefore, it was
considered necessary to include the asperity height distribution in the revised model
(PERC II). Due to constant asperity height, only an average indentation depth was
derived in the previous model (PERC I) to indicate surface roughness of the polished
wafer. When asperity height distribution is used, asperities at different heights lead to
different indentation depth, resulting in information about distribution of scratches
(indentation depths).This data can be used to estimate the surface roughness as root mean
square roughness (RMS).
To incorporate the asperity distribution, Gaussian distribution of asperities is
assumed, as the reported literature cites Gaussian distribution [116]. The pressure at
contacting asperities were calculated individually to estimate the indentation depth of
trapped particles.
Abrasive Particle Size Distribution
The role of slurry in CMP is two-fold. The chemicals in the slurry are needed to
react with wafer surface resulting in formation of a surface layer which is amenable to
enhanced material removal. The abrasives in the slurry on the other hand, remove this
modified surface layer causing material removal. The effect of chemicals is attributed to
surface layer of different mechanical properties in PERC I model. Whereas, the effect of
abrasives is considered by assuming average size of the particles. The monosized slurry
assumed in the PERC I model is an ideal case, since real CMP slurries are seldom
monosized (Fig. 3-15). Oxide polishing slurries are commonly purchased in concentrated
form and diluted on-site to minimize shipping and labor costs. Some mult-component
tungsten and copper polishing slurries must be combined shortly before use because of
their short postmix lifetime. The slurry stability can change after it is combined with the
oxidizer or due to dilution leading to polydispersity.
0) *
O s .. -/ ... \ ..+- .
0 60 T7 80 90 100
Diameter (nm)
Fig. 3-15. Particle size distribution of silica slurry [118]
Sometimes mixed abrasive slurries are used [32], which invariably exhibit
polydisperse particle size distribution. To account for the particle size distribution effect,
PERC I model was modified to incorporate slurry abrasives with Gaussian size
distribution
PERC II Model Development
Conceptually, the difference between the PERC I and PERC II models can be
illustrated as shown in Figs. 3-16 and 3-17. The PERC I model assumes spherical pad
asperities of equal height all over the pad (Fig. 3-16a). As the slurry is assumed to be
76
monosized, each asperity traps equal number of particles in CMP resulting in equal
removal rate by each asperity and each particle(Fig. 3-16b).
Wafer
Pad aspriilics
. .
\/
Pad asperities
(b)
Fig. 3-16. Schematic of pad, slurry and wafer in PERC I model. (a)Before contact. (b)
During CMP.
Whereas in the PERC II model (Fig. 3-17a), the spherical pad asperities are
assumed to have a height distribution as observed experimentally [116]. As the slurry is
assumed to be polydisperse, each asperity traps different number of particles of different
particle sizes in CMP according to pad asperity height and particle distribution (Fig. 3-
17b). Thus the PERC II model includes realistic contributions of pad and wafer
characteristics compared to the PERC I model. The model formulation is detailed in the
following section.
..........
...............
................
................
................
................
................
.................
..............
. . . .
.............
...........
.........
............
. . . . .
. . . . .
- - - - -
. . . . . .
. . . . .
. . . . . .
. . . . . . .
Wafer
'--hury particles
. . .
\
Pad asperities
Wafer
/. ... ... ... ..
. . . . .
Fig. 3-17. Schematic of pad, slurry and wafer in PERC II model. (a) Before contact.
(b)During CMP
Real contact area (Areal) between pad and wafer due to applied load (L) is given by:
Ar =rNA, (z h)(z)dz
L 4NAoE3/2 z -h)32(z)dz
P(z) = exp
a =~
(3.1)
(3.2)
(3.3)
2
z
2a2
where Nis the number density of the pad asperities (i.e., number of asperities per unit
area), Ao is the total area of the flat wafer surface, and h is the distance between wafer
. . .
. . . .
Wafer
.......... ..
..............
...............
................
.................
..................
...................
...................
...................
....................
................
.................
surface and the pad asperity average plane under the given conditions. E* is the
composite modulus between the two surfaces in contact, defined as
1 1-v2 1v (34)
-- + W (3.4)
E* Ep E
where, E and v are the Young's modulus and the Poisson's ratio, and the subscripts p and
w represent the pad and the wafer, respectively. Equations (3.1) and (3.2) are derived
from Hertz theory as in Greenwood-Williamson model [77, 117], yielding contact area
and load at a given asperity of height 'z' to be:
Areal(Z)=7 P (z-h) (3.5)
L(z)=(4/3)E* 1/ 2(z-h)3/2 (3.6)
where 'h' is calculated from Eq. 3.2, since applied load and pad properties are known.
The underlying assumption for Eq. 3.2 is that the applied normal force on the wafer is
supported entirely by the pad asperities. As was mentioned previously, experimental
measurements indicate that the hydrodynamic pressure is negligible under typical CMP
operational conditions [69].
From equations 3.5 and 3.6, contact pressure and contact area at a given asperity
height can be calculated assuming that the pad asperity distribution is known. The pad
asperity distribution can be determined from routine surface roughness measurement
techniques like optical and force profilometers or atomic force microscopy (AFM).
Contact pressure according to Hertz theory, on an asperity of height z is given by:
P(z) =4E* (z- h)2 (3.7)
3z-f1i
35c/72
Plastic Deformation of Wafer Surface
When an abrasive particle is indented into the wafer surface, elastic and plastic
deformation of wafer and abrasive particle occur. However, in CMP, slurries are
designed such that the chemistry changes the modulus of the wafer surface for enhanced
material removal. In the current model, particles are assumed to be rigid such that they do
not deform during polishing. When the indentation of abrasive particle is beyond the
modified surface layer, the substrate indentation could yield either elastic or plastic
response depending on the substrate, for example in case of silica, the deformation will
still remain plastic whereas for copper, it would be elastic deformation followed by
plastic deformation. Therefore, in material removal calculation, indentation into ductile
wafers is not considered to cause material removal, as it leads to change in shape only,
whereas in the case of brittle material like silica, cracks are developed leading to material
removal.
Slurry Characteristics Particle Size Distribution
The abrasive particles in the slurry are assumed to be spherical in shape with a
given size distribution y(r).As mentioned earlier, particle size distribution Y(r) was
assumed to be Gaussian, given by:-
S(r) ex (r -Ur)2 (3.8)
ao. [2; 2o.2
where [Lr is the mean particle radius and Or the standard deviation.
Material Removal by Single Particle
The overall material removal rate (RR) is assumed to be the multiple of the
removal rate by a single particle and the number of particles trapped between pad
asperities and the wafer:
RR = (1/Ao)J In(r,z) RR,(r,z)drdz (3.9)
where n(r,z) is the number of slurry particles anchored between pad asperity and the
wafer and RR,(r,z) is the volumetric removal rate by a single particle of radius r trapped
by asperity of height z. The overall removal rate, RR is typically described in terms of the
thickness of the wafer surface layer removed per unit time; hence the division by the
nominal wafer area Ao.
Schematic illustration of single particle abrasion is given in Fig. 3-18. The material
removal is viewed as a sliding-indentation process in which the volumetric removal rate
by a single abrasive particle is given by
RR,(r,z)= A,(r,z)V (3.10)
Where A,(r,z) is the cross-sectional area of the particle indented into the wafer
under pressure (hashed area in Fig. 3-18), and Vis the relative velocity between the pad
asperity and the wafer.
I T
Sb Pad
t, Pad t
F. Pad F.
Fig. 3-18 Schematic of single particle abrasion showing the removal of chemically
modified surface layer, (a). co < t; (b). co > t with soft wafer materials.[72]
Assuming that the hard particle maintains its spherical shape,
A, (r,z) = -a(r,z)(r o(r,z))+ r2 arcsin a(r,z) (3.11)
r
where co(r,z) is the indentation depth and a(r,z) is the radius of the circular indentation on
the wafer surface caused by the particle of radius r trapped under asperity of height z.
Because co(r,z) is very small compared to r, a(r,z)/r is also very small and Equation 3.11
can be linearized about a(r,z)/r as already discussed:
Az (r, a(r, z)(r, z) )+ ( (3.12)
3 r
Furthermore, since a2(r,z)=(2rco(r,z)-co2(r,z)) and co(r,z) is much smaller than r,
Equation 3.12 can be rewritten as
A4 (r, z) 2a3(r,z) (3.13)
3r
Once the indentation depth, co(r,z) (or a(r,z)) is known, the volumetric removal rate
by a single particle can be calculated from Eq. 3.13. It should be noted that co and a are
dependant on particle radius and height of the trapping asperity. Removal rate per particle
is given by:
RR, (r, z) = 2a(rz) V (3.14)
3r
Depending on the relative magnitude of the indentation depth (co(r,z)) as compared
to the surface layer thickness (t), the material removal may be classified into two
different regimes as depicted in Fig. 3-18: (a) o(r,z) < t and (b) o(r,z) > t and can be
calculated using force balance as follows.
Considering that the pad is sufficiently hard as compared to the modified layer, the
value of b in Fig. 3-18 can be close to the particle radius R,
( -1/2
P(z)
a(r,z) = r i when co(r,z) < t (3.15)
and
a(r, z) = r2p(z) + 2(BW Be)rt when co(r,z) > t (3.16)
BI
where, Be is the effective Brinell hardness of the chemically modified surface layer. In
the following section, the number of particles in contact with the wafer surface is
estimated to determine the overall removal rate.
Number of Active Abrasive Particles
When the wafer is pressed against the pad and is in sliding motion relative to the
pad, the slurry particles can be entrapped only in the region where the local distance
between the wafer and the pad asperity is smaller than the particle diameter, 2R. For a
slurry consisting of particles with Gaussian distribution, mean particle size (Rm) is used
to calculate the number of particles that can be trapped on the pad asperity of height z:
N(z) 2Rm -Areai(z) p (3.17)
where Rm is the mean abrasive particle radius and Areai(z) is the real contact area between
the wafer surface and the pad asperity as given by eq. 3.5. p, is the number density of the
particles in the slurry (i.e., number of particles per unit slurry volume), that can be
calculated from the solid loading Ca (wt%) of the slurry particles as
3C, p,
Pn ,= 3a 0p (3.18)
4;r pr3 (r (r)dr
where pa and ps are the densities of the particle and the slurry solution, respectively and
y(r) the slurry particle size distribution. Substitution of Eq. 3.18 into 3.17 gives
3 Rm Caps Area, (z)
N(z) 2=( (3.19)
2 ap, frv ,(r)dr
From eq. (3.19), the number of trapped particles of radius R is given by,
n(R,z)=N(z)*y(R)= 3 Rm C Area (z) (r) (3.20)
2 ;Tr fJr3 V(r)dr
Estimation of the Overall Removal Rate
The overall material removal rate can be determined by following these steps:
h is calculated using eq. 3.2.
Contact pressure P(z) at a given asperity is calculated along with the real
contact area Areai(z) of that asperity with wafer.
Total number of particles N(z), that are trapped are determined, from which
number of particles of given radii, n(r,z) are derived.
From P(z), removal rate by single abrasive of radius r is calculated which is
multiplied with n(r,z) to give removal rate by that particle size. This process
is repeated to account for all the particles of different radii trapped on that
asperity yielding material removal rate by particles trapped on a given
asperity.
It should be noted that there are 4(z)NAo asperities at a given height which
will each contribute the same material removal.
The above step is repeated for all the pad asperities of various heights that
are in contact with wafer (determined by h) to give overall removal rate by
the following equation:
RR = (1/Ao)J In(r,z) RR,(r,z)drdz
-RR= CpN Area(z) ) w(r)(r,z ddz (3.21)
p, r3 y(r) dr h r
r
where, a(r,z) varies as given in eqs.3.15 and 3.16 depending on particle radius and
asperity height.
The indentation depth, co(r,z) is used to calculate the surface roughness RMS of the
polished wafer according to the following equation:
0)2
RMS = (3.22)
n
where coi is the indentation depth.
The overall removal rate predicted by Eq. 3.21 and surface roughness given by
Eq. 3.22 depend on various parameters that represent the geometry of the pad and the
particle (i.e., 3, o and R), material properties (E*, Bw and Be), thickness (t) of the modified
surface layer on wafer, and the relative velocity of the pad (thus the particle) to the wafer
(V). The PERC II model will be validated with respect to critical CMP parameters in the
next chapter.
CHAPTER 4
MODEL ASSESSMENT
The parameters governing the material removal calculation consist of CMP
operating parameters, slurry parameters, pad and wafer parameters, as listed in Table 4-1.
However, some of these parameters, especially the sub-nanometer thick modified layer
hardness and thickness can not be determined experimentally due to equipment
limitations. The remaining parameters are relatively easy to investigate experimentally
for many CMP systems. Hence, modified layer hardness and thickness are used as the
fitting parameters in the model.
Table 4-1. Parameters included in the revised PERC model.
Operating Parameters Pad properties
Pressure Pad modulus
Velocity Asperity height distribution
Asperity radius
Slurry parameters Asperity density
Particle size distribution
Solids loading Wafer properties
Abrasive density Modified layer thickness
Solvent density Modified layer hardness
Wafer hardness
Model Validation
In this section, PERC II model predictions are compared with experimentally
measured values to assess the reliability of the proposed CMP model. The experimental
literature data is also compared with PERC I predictions to assess the advantage of PERC
II over PERC I. Though PERC II is a refined model compared to PERC I, the model is
still a simplified version of real CMP, mainly in estimating the number of trapped
particles and indentation depth considering an abrupt transition between modified layer
and wafer substrate. The removal rate values for both models are normalized by a
constant to have reasonable modified layer thickness values. Surface roughness data is as
produced by the models without any normalizing factor. The size distribution of slurries
given in the respective literature are fit to the closest Gaussian distribution, which is used
in the model.
Silica CMP: Micron Sized Slurries
The polishing results of silica by silica abrasives as reported by Choi et al. [131]
are considered here. Micron sized CMP slurries were formulated by dispersing sol-gel
silica particles of different sizes from Geltech Corporation in deionized water and
adjusting pH values to 10.5. The average diameter of the slurry particles were 0.2, 0.5,
1.0 and 1.5 itm. The size distribution of the slurries is plotted in Fig. 4-1. A 2 |tm thick
silica layer was deposited on p-type silicon wafer by plasma-enhanced chemical vapor
deposition. 1.5 x 1.5 in silica samples were used to mimic the wafers. A Struers RotoPol-
31 tabletop polisher was used with IC 1000/Suba IV stacked pad. The following
conditions were applied: down pressure of 3.5 psi, slurry flow rate of 100 ml/min, and
rotation speed of 150 rpm (110 cm/s). The polishing rates were determined by measuring
oxide film thickness using variable angle spectrometry ellipsometry and surface quality
was assessed by atomic force microscopy (AFM).
The polishing rate was investigated as a function of particle size and solids loading.
The removal rate decreased with increasing particle size (Fig. 4-2). The parameters used
are: wafer Hardness=8 GPa, modified layer Hardness = 450 MPa, modified layer
thickness = 7.2 A and Normalization factor = 100. The PERC II model correlates well
with the experimental values than the PERC I model. The reported literature belongs to
the substrate indentation regime. The model projects that enhanced removal rate can be
obtained by using smaller sized slurries. The surface roughness on the other hand,
increases with particle size on an average for all the slurries considered (Fig. 4-2b). This
is not surprising as the larger size particles have deeper indentation depths increasing
surface roughness. Both the models predict the average trend in surface roughness with
increasing particle size.
25
-- 0.2pm
20 0.5pm
S---- 1.0pm
o 15 1.54n
E -O
0
0 1 2
Particle Size (mn)
Fig. 4-1. Slurries used in oxide polishing [131].
Silica CMP: Nanosilica slurries
The slurry sizes in the earlier study spanned from 0.2 to 1.5 jtm, which is well
above the normal slurry size range (<300 nm). Considering the nanosize slurries of 50-
150nm mean particle sizes, CMP behavior of silica polishing by silica abrasives is
investigated further [116]. Slurries with mean particle diameters of 50, 80, 110 and
140nm were used to polish 100mm diameter, p-type polished silicon wafers that
contained a thickness of 1.3 pm thermally grown oxide. The silica particles were
precipitated from a colloidal solution. The polishing was done at three different levels of
pressure and relative pad-wafer velocities: 18kPa, 0.4m/s; 32kPa, 0.6m/s; and 45kPa,
0.9m/s. The parameters used in the model are: wafer Hardness=8 GPa, modified layer
Hardness = 250 MPa, and modified layer thickness = 9.3 A. The removal rates for the
given set of pressure and velocity with different particle sizes is shown in Fig. 4-3. The
material removal rates with slurry mean sizes of 10 and 20nm were ignored, as the
authors [116] noticed considerable particle staining on the polished surfaces indicating
agglomeration.
The PERC II model corresponds with the experimental results better than the
previous PERC I model. Each set of modeled removal rates for given pressure and
velocity were normalized individually with normalizing factor ranging from 185-200.
The discrepancy in the removal rate at particle size 50nm between experimentally
observed values and modeled values can be attributed to agglomeration, which is
suggested by the authors for that size [116].
250
Experimental
200 1 PERC II Model
---PERCI Model
150
5 100
E
50
0 i i
0 500 1000 1500 2000
Mean Particle Size (nm)
(a)
Fig. 4-2. Silica polishing by slurries in the micron range. (a) Removal rate. (b) Surface
quality
|
Full Text |
PAGE 1
PARTICLE SCALE MODELING OF MATERIAL REMOVAL AND SURFACE ROUGHNESS IN CHEMICAL MECHANICAL POLISHING By SURESH BABU YERUVA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2005
PAGE 2
Copyright 2005 by Suresh Babu Yeruva
PAGE 3
This document is dedicated to Sri Parama hansa Yogananda, my parents and sister.
PAGE 4
iv ACKNOWLEDGMENTS I would like to express my gratitude a nd appreciation to my advisor, Dr. Brij Moudgil, for his valuable guidance and support throughout this study. His sincere dedication towards his students encouraged me to perform my best and expand my potential. Sincere thanks are also due to Dr Chang-Won Park, for his valuable guidance and support throughout the study. I would also like to acknowledge the other members of my advisory committee, Dr. Rajiv Singh, Dr. Wolfgang Sigmund, Dr. Hassan El-Shall, Dr. Susan Sinnott and Dr. Dinesh Shah, for thei r time, helpful discussions and guidance. I would like to thank Dr. Yakov Rabinovich fo r help with atomic force microscope (AFM) and Fourier transform in frared spectroscopy (FTIR) an alysis, and Gerald Bourne from Major Analytical Instrumentation Center (MAIC) for help with focused ion beam (FIB) and nanoindentation data. Furthermore, I sincerely acknowledge the guidance and support given by Kuide Qin at Praxair Inc. I gratefully acknowledge th e National Science Foundatio ns Particle Engineering Research Center (PERC) for financially suppor ting this research. Special thanks are due to PERC staff members, Gary Schieffele Gill Brubaker, Mike Beasley, and John Henderson who made this work possible. My colleagues, Bahar Basim, Jeremiah Abiade, Su-Ho Jung, Vijay Krishna, Madhavan Esayanur, Scott Brown, Parvesh Sharma, Kyoung-Ho Bu, Pankaj Singh, Debamitra Dutta and Ivan Vakarelski, who ha ve contributed to this research by their valuable discussions and friendship, are also gratefully acknowledged.
PAGE 5
v TABLE OF CONTENTS Page LIST OF TABLES............................................................................................................vii LIST OF FIGURES.........................................................................................................viii ABSTRACT xi CHAPTER 1 INTRODUCTION........................................................................................................1 Multilevel Me tallization...............................................................................................2 Need for Planarization..................................................................................................4 Planarization Techniques..............................................................................................5 Overview of the Chemical M echanical Polishing Process...........................................7 History...................................................................................................................7 Advantages and Disadvantages of Ch emical Mechanical Planarization...............8 General CMP applications.....................................................................................9 Cmp Output Parameters........................................................................................9 Removal rate................................................................................................10 Global planarization.....................................................................................12 Selectivity.....................................................................................................13 Measurement Techniques....................................................................................14 End Point Detection.............................................................................................15 CMP Consumables.....................................................................................................16 CMP Slurry..........................................................................................................18 Rate of surface reaction................................................................................18 Slurry agglomeration....................................................................................19 Chemical Mechanical Polishing Pads.................................................................20 Effect of pad grooves...................................................................................23 Physical and chemical cha nges in pad during polishing..............................23 Effect of temperature change on polishing pad............................................26 Non-uniformity of the polishing pad............................................................27 CMP process in the future..........................................................................................27 Scope of Dissertation..................................................................................................28 2 CMP MODELING: OVERVIEW OF PERC I...........................................................30 Literature Review.......................................................................................................31
PAGE 6
vi Preston Equation..................................................................................................31 Modifications to Preston Equation......................................................................34 Slurry Erosion......................................................................................................36 Pad Considerations in CMP Models....................................................................39 Slurry Considerations in CMP Models................................................................42 PERC I CMP Model...................................................................................................44 Model Development............................................................................................45 Estimation of the Overall Removal Rate.............................................................54 3 PERC II CMP MODEL..............................................................................................59 Assumption Validation...............................................................................................59 Pad-wafer Contact Area......................................................................................59 Optical microscopy......................................................................................61 Fourier transform infrar ed spectroscopy (FTIR)..........................................64 Pad Asperity Height Distribution........................................................................73 Abrasive Particle Size Distribution.....................................................................74 PERC II Model Development.....................................................................................75 Plastic Deformation of Wafer Surface................................................................79 Slurry Characteristics Part icle Size Distribution..............................................79 Material Removal by Single Particle...................................................................79 Number of Active Ab rasive Particles..................................................................82 Estimation of the Overall Removal Rate....................................................................83 4 MODEL ASSESSMENT............................................................................................85 Model Validation........................................................................................................85 Silica CMP: Micron Sized Slurries.....................................................................86 Silica CMP: Nanosilica slurries...........................................................................87 Copper CMP........................................................................................................91 Tungsten CMP.....................................................................................................92 Model Predictions.......................................................................................................94 Removal Rate as a Function of Partic le Size for a Given Solids Loading:.........95 Effect of Modified Layer Thickness....................................................................96 Effect of Modified Layer Hardness.....................................................................99 5 SUMMARY AND SUGGESTIONS FOR FUTURE WORK.................................101 LIST OF REFERENCES .................................................................................................106 BIOGRAPHICAL SKETCH........................................................................................... 118
PAGE 7
vii LIST OF TABLES Table page 1-1. Interconnection delay (RC) in silicon VLSI chip.........................................................3 1-2. Chemical mechanical polishing applications.............................................................10 4-1. Parameters included in the revised PERC model.......................................................85 4-2. Parameter values used for comparing PERC I and II models....................................95 5-1. Process parameter ranges used in sensitivity analysis..............................................103
PAGE 8
viii LIST OF FIGURES Figure page 1-1. Future trends in interconnects ......................................................................................4 1-2. Schematic of a device fabricated using multilevel metallization.................................5 1-3. Different degrees of local and global surface planarity................................................6 1-4. Schematic of chemical mechanical polishing process..................................................7 1-5. Surface layer formation on Cu w ith various slurry chemistries.................................11 1-6. Micro and nanoscale phenomena during CMP ..........................................................13 1-7. Parameters involved in CMP process.........................................................................17 1-8. Chronoamperometry meas urements for tungsten ......................................................19 1-9. Type I CMP pad.........................................................................................................22 1-10. Type II CMP pad......................................................................................................22 1-11 The absorbance spectra of used a nd dry IC 1000 pad along with dried silica slurry.........................................................................................................................25 2-1. Polished wafer surfaces showing nanoscratch surfaces.............................................33 2-2 Pad consideration in PERC I.......................................................................................46 2-3. Schematic of single particle abra sion showing the removal of chemically modified surface layer..............................................................................................51 2-4. Schematic of PERC I predictions for increasing particle size....................................57 2-5. Polishing of silica by nanoslurr y showing optimal particle size................................57 2-6. Polishing of silica at various pressure, velocity and particle sizes showing optimal particle size...............................................................................................................58 3-1. Schematic of the fixture used in measuring pad-wafer (dry) contact area.................62
PAGE 9
ix 3-2. Pad asperity contact spots s een through optical microscope......................................63 3-3. Pad asperity contact area dependence on applied load. Each legend represents a selected asperity contact...........................................................................................63 3-4. Pad contact area as a f unction of applied load............................................................64 3-5 Schematic of the FTIR-ATR technique used to measure padwafer contact area.......65 3-6. Analysis of the FTIR-ATR spectra.............................................................................67 3-7. Absorbance spectra of pad (in pH 4) at different applied pressures...........................67 3-8. Contact area variation with applie d load in dry and wet conditions..........................68 3-9. Contact area in dry and wet sy stems in various pH conditions..................................68 3-10. Colloidal probe AFM tip..........................................................................................70 3-11. Force-distance curve indicating softer modified layer on the pad surface...........71 3-12. Force-indentation curves from AFM for different conditions..................................72 3-13. Schematic of pad surface as deduced from AFM data.............................................72 3-14. Schematic of polishing pads.....................................................................................73 3-15. Particle size distribu tion of silica slurry...................................................................75 3-16. Schematic of pad, slurry and wafer in PERC I model..............................................76 3-17. Schematic of pad, slurry and wafer in PERC II model............................................77 4-1. Slurries used in oxide polishing..................................................................................87 4-2. Silica polishing by slurri es in the micron range.........................................................88 4-3. Silica polishing by colloidal nanosilica......................................................................89 4-4. Silica polishing by nanosized slurries. .....................................................................90 4-5 Copper CMP as a functi on of particle size..................................................................92 4-6. Alumina slurry si ze distributions................................................................................93 4-7. Effect of particle size on tungsten CMP.....................................................................94 4-8. Comparison between PERC I and PERC II mode ls with constant solids loading of different particle sizes..............................................................................................96
PAGE 10
x 4-9. Model comparison with varyin g modified layer thickness........................................98 4-10. CMP performance comparison with varying modified layer hardness..................100 5-1. Sensitivity analysis of silica CMP system................................................................103
PAGE 11
xi Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PARTICLE SCALE MODELING OF MATERIAL REMOVAL AND SURFACE ROUGHNESS IN CHEMICAL MECHANICAL POLISHING By Suresh Babu Yeruva December 2005 Chair: Brij M. Moudgil Cochair: Chang-Won Park Major Department: Materials Science and Engineering Chemical mechanical polis hing (CMP) is widely adopt ed in producing excellent local and global planarization of microelectronic devices. It has been demonstrated experimentally that the polishing performance is a result of the syne rgistic effect of both the chemicals and the particle s involved in CMP. However, the fundamental mechanisms of material removal and the in teractions of the chemical and mechanical effects are not well understood. A comprehensive model for CM P was developed taking into account both the chemical and mechanical effects for slurries with a given particle size distribution. The model developed (PERC II) is based on a previously developed model (PERC I). The chemical aspect is attributed to the chemical modification of the surface layer due to slurry chemistry, whereas th e mechanical aspect is incorporated by indentation of particles into the modified layer and the substr ate depending on the
PAGE 12
xii operating conditions. In this study, the eff ects of particle size and pad asperity distributions are included in the model. The contact area of pad with wafer was measured in dry and wet conditions in different pH solutions using optical mi croscopy and Fourier transform infrared spectroscopy respectively. Pad surface mechanical properties in dry and wet states were also investigated using atomic force micr oscopy. The contact area results obtained were utilized in modeling to estim ate the pad modulus leading to pad-wafer contact stress distribution. The predictions of the mode l show a reasonable agreement with the experimental data. The model is validated for oxide a nd metal CMP systems. The PERC II model not only predicts the overall removal rate, but also the surface roughness of the polished wafer in selected systems. The developed model can be used to optimize the current CMP systems and provide insights into future CMP endeavors.
PAGE 13
1 CHAPTER 1 INTRODUCTION The past three decades have witnessed rapid growth of the electronics industry and its impact on human lifestyle. Ultra-largescale integrated (ULS I) circuits, with 108 or more devices on a chip, can now be fabricat ed on semiconductor wafers, reducing cost and increasing the performance of electroni c products. The number of components on a memory chip has approximately doubled ever y two years over the past two decades, following the famous Moores law [1]. Th e advancement in fabrication technology shrinks the minimum dimension of the device -feature by about 13% per year. This trend is expected to continue in the future even as 65 and 45 nm processes are set for production in 2005 [2] and near future. The relentless competitor and customer driven demand for increased circuit density, functionality and versat ility has led to revolutionary advances in the front end of the chip manufacturing line where the circuit elements ar e fabricated, and the back end where these elements are appropriately wi red within the integrat ed circuit (IC) [3]. Chip interconnections, or interconnects, serve as local and gl obal wiring, connecting circuit elements and distributing power [4]. To incorporate and accommodate the improvements such as decreased feature size, increased device speed and more intricate designs, research in the back end of line (BEOL) processes has become equally important as the development of the front end of line (FEOL) processes to reduce gate oxide thickness and channel length.
PAGE 14
2 While recent path breaking innovations [5-7] in the field of lithography and patterning have brought progressive de vice scaling, the development of a planar BEOL approach, which incorporates the use of chemical mechanical polishing (CMP) to planarize inter-level dielectrics and meta l stud levels, represents a significant advancement in BEOL processes. Innovation in BEOL technology is required in each technology generation, since only part of the de nsity increase could be achieved with improvements in lithography. Multilevel Metallization In earlier stages, the complementary metal oxide semiconductor (CMOS) device structure had multiple isolated devices conne cted by a single level of interconnect wiring. Scaling down of the devices was very effectiv e in achieving the goals of increased device density, functional complexity and performa nce. However, device scaling became less profitable, and speed and complexity we re dependant on the characteristics of interconnects that wired them [8]. With the single level metallization scheme, the total area occupied by the wiring on the chip signifi cantly increased with the increase in the active density on the chip. The interconnect delay is usually measured as RC time delay, which is the total time taken by the voltage at one end of the me tal line to reach 63% of the total value of the step input applied at the other end. Th is reduction is due to resistance of the interconnect wiring metal (R) and the interlay er dielectric capacitance (C) [9]. The resistance of the line is given by R = l/wd (1.1) where is the resistivity of the wiring material l, w, and d are the length, width, and thickness of the wiring material respectivel y. The capacitance of the line is given by
PAGE 15
3 C = wl/t (1.2) where is the relative permittivity and t is the thickness of the insulator. The total RC delay can be given by RC = l/wd wl/t = l2/td (1.3) Thus, it can be seen from Eq. (1.3) that RC delay is independent of the line width and further scaling of line width translates into reduction of IC line thickness which in turn increases the RC delay. Othe r factors such as parasitic capacitances and cross-talk become dominant for sub 0.5 m integrated circuits. One approach to decrease the RC delay is by incorporating metals of low resist ivity, and interlayer dielectrics of lower dielectric constant. Table 1-1 gives the simple RC time constants calculated for a few metals of given Rs (sheet resistance) and 1 mm length on 1 m thick SiO2 [9] The other approach implemented widely in the industr y is forming multileve l metallization schemes where different levels of metal interconnections are isolated by dielectrics and are connected by vertical vias. Table 1-1. Interconnection delay (RC) in silicon VLSI chip. Metal Bulk resistivity ( cm) Polycrystalline film resistivity ( cm) Film thickness () Rs ( /square) Delayi (ps/mm) Polysilicon ~1000 5000 20 690 CoSi2 10 15 2500 0.6 21 MoSi2 ~35 ~100 2500 4 138 TaSi2 45 55 2500 2.2 76 TiSi2 13 15 2500 0.6 21 WSi2 ~25 70 2500 2.8 97 W 5.65 8-10 2500 0.32-0.4 11-14 Mo 5.2 8-10 2500 0.32-0.4 11-14 Al 2.65 2.7 2500 0.11 4 Cu 1.67 2.0 2500 0.08 3 iDelay=RC=34.5 Rs (ps/mm) for 1 mm length conductor on 1 m thick SiO2 [10].
PAGE 16
4 The increase in the levels of the metallization lines implies that packing density need not keep pace with the device density and the minimum metal line feature does not have to scale at the same pace as the gate width. The foremost reason behind the implementation of multilevel metallization sc hemes is the reduction in the length of the metal lines, thereby reducing the RC delay sizably ( Eq. 1.3 ), and allowing direct routing of the active devices. In places where metal wiring length cannot be reduced, routing can be done at the upper levels without reducing the metal line width, thus reducing the RC delay due to the higher surface area. Starting wi th two levels of me tallization, the levels of metallization have increased up to 8 by 2001 [11]. The future trends in the levels of metallization can be seen in Fig. 1-1 [12]. Fig. 1-1. Future trends in interconnects [12] Need for Planarization Deposition of multiple layers leads to conformation of deposited layer on the substrate features or already deposited pattern. When multilevel metallization is carried out without any planarization step, non-planarized surface topography (Fig. 1-2a) results
PAGE 17
5 in several processing difficulties. The irregul ar surface causes a hindrance in conformal coating of the photoresist and efficient pattern transfer with contact lithography. The anomalies in the surface cause the variation of the thickness in fine line widths (sub 0.5 m) depending on photoresist thickness. Fig. 1-2 shows a comparison between planarized and non-planarized surface topography. Effectivel y planarized surfaces have a number of benefits such as (a) higher photolithography an d dry etch yields, (b) elimination of step coverage concerns, (c) minimi zation of prior level defects, (d) elimination of contact interruption, undesired contact s and electro-migration eff ects, (e) reduction of high contact resistance and inhomogeneous metallization layer thickness, and (f) limitation in the stacking height of metalliz ation layers. Thus the demand for developing devices with enhanced performance nece ssitates planarization. Fig. 1-2. Schematic of a device fabricated using multilevel meta llization. (a) Nonplanarized structure. (b) Pl anarized structure [8, 13] Planarization Techniques Techniques such a spin on deposition (S OD), reflow of boron phosphorous silicate glass (BPSG), spin etch plan arization (SEP), reactive ion etching and etch back (RIE EB), SOD + EB and chemical mechanical polishing (or chemical mechanical planarization) (CMP) are the most prominent of several competing technologies available
PAGE 18
6 to achieve local and global planarization (Fi g. 1-3). The plasma enhanced chemical vapor deposited oxides have limited capability of gap filling and are restricted in their gap filling ability below patterns having 0.3m feature size. High-density plasma deposited oxides have acceptable gap filling capabiliti es; however, they produce variation in surface topography at local as well as global level. Even though spin-on deposited (SOD) doped and undoped oxides and polymeric materi als have acceptable ability for gap filling, CMP is the only technique whic h produces excellent local and global planarization of microe lectronic materials. Fig. 1-3. Different degrees of local and global surface planarity [14].
PAGE 19
7 Overview of the Chemical Mechanical Polishing Process Current semiconductor fabrication tec hnology for logic and memory devices requires CMP to achieve the required multilevel interconnection densities. Indeed, each silicon wafer can be exposed to 15 or more CMP steps before final device assembly. A schematic diagram of the CMP process is shown in Fig. 1-4 In CMP, the wafer is pressed face down against a rotating polishing pad, while a chemical and physically abrasive slurry fed between pa d and wafer planarizes the wafe r. The material removal in simplest terms takes place by synergistic mechanism between chemical reaction on the wafer surface due to the slurry, and physical re moval of the reacted material by the slurry abrasives. The various mechanisms between pa d, slurry and wafer are tailored to enhance material removal, planariza tion, surface finish and higher selectivity compared to the underneath layer. Fig. 1-4. Schematic of chemical mechanical polishing process. History The modern day application of the CMP pr ocess in the semiconductor industry was for polishing the raw silicon wafers to ach ieve a global flatness. One of the most widespread applications of CMP outside th e semiconductor industry is in optical lens
PAGE 20
8 polishing. In fact, the first semiconductor CM P machine was an innovation of the lens polishing industry. The first wafers polished us ing CMP were commercially available in the early 1960s [15]. There was wide spread industrial implementation of the different variants of the CMP by companies such as NEC, National Semiconductor, Hitachi, etc. [10]. This led to the introduction of the firs t commercial polisher designed specifically for CMP by Cybeg, Japan in 1988. Later, Interna tional Sematech identified CMP as a technology critical for the future of IC ma nufacturing and launched a project to develop competitive and advanced CMP tools in th e US [16]. Once the technology and required equipment were available, CMP was used beyond polishing of inter-level dielectric layers to polishing of tungsten, which was deposited in via openings between metal layers. Copper was not used for interconnects until th e availability of CMP even though it has lower resistivity than aluminum [17], as it could not easily be etched by reactive ion etching. This enabling polishing technology has sp read to several othe r applications, the significant one being in shallow trench isolation (STI) [18]. Advantages and Disadvantages of Ch emical Mechanical Planarization The advantages of CMP are (a) it achieves global planarization, (b ) a wide range of materials can be planarized, sometimes in th e same polishing step, (c) it reduces severe topography to allow fabrication with tighter design rules, (d) it provides an alternate means of patterning metal, eliminating the need to plasma etch, (e) it improves metal step coverage due to reduction in topography a nd (f) it does not use hazardous gasses common in dry etch processes. Conversely, the disadvantages of the CM P process are (a) poo r control over the process with narrow latitude in process vari ables, (b) new types of defects introduced by the CMP process itself (scratch es, particle residue, etc.), (c) difficult end-point detection
PAGE 21
9 for desired thickness and (d) expensive to operate because of costly equipment and consumables. General CMP applications The process of CMP was initially develope d and implemented for planarization of SiO2 which is used as an interlayer dielec tric in a multilevel metallization scheme. Tungsten is used as an interconnect plug to the source, drain, and gates of transistors in Si microprocessor chips. Initially Ti and Ti N barrier layers are deposited, followed by chemical vapor deposition of W to fill the contact vias. Proceeding from achieving local and global planarization of SiO2, removal of excessive tung sten from the horizontal surfaces on the wafer pattern proved to be an asset for subsequent Al metallization. Hence CMP was developed with a two-fold a pproach of (a) planar izing oxide and (b) removing the via fill metal from the horizontal surfaces. Along with its successful implementation for the achievement of th e above-mentioned objectives, CMP is now extended to (a) polishing of different metals like Al, Cu, Pt, Au, Ti, Ta, etc., (b) polishing of different insulators like SiO2, Si3N4, various low-k dielectrics, doped and undoped oxides of silicon, (c) polysilicon, (d) ceramic s like SiC, TiN, TaN, etc., (e) multichip modules, (f) packaging; (g) optoelectronic components, (h) flat panel displays, (i) microelectromechanical systems (MEMS), and (j) magnetic recording heads and CD read write drives [19]. The major applic ations of CMP are summarized in Table 1-2 Cmp Output Parameters The critical output parameters of CM P are described below. Depending on the application, different output parameters are optimized.
PAGE 22
10 Table 1-2. Chemical mechanic al polishing applications. Materials Application Metal Al Cu Ta Ti TiN, TiNxCy W Dielectric Cu-alloys Al-alloys Polysilicon SiO2 BPSG PSG Polymers Si3N4 or SiOxNy Other Aerogels ITO High k dielectrics High Tc superconductors Optoelectronic materials Plastics, ceramics Silicon on Insulator (SOI) Interconnection Interconnection Diffusion barrier/adhesion promoter Diffusion barrier/adhesion promoter Diffusion barrier/adhesion promoter Interconnection e-emitter Interconnection Interconnection Gate/interconnect ILD ILD ILD ILD Passivation layer, hard mask ILD Flat panel Packaging Interconnection/packaging Optoelectronics Packaging Advanced device/circuits Ref. [19] Removal rate The polishing steps involving CMP of STI and metal mainly emphasize removing an overburden of one material and stop on another material leaving a planar surface. A high material removal rate is one of the ke y outputs desired from such CMP processes. The removal rate of the material being polishe d is strongly influenced by slurry chemistry and pressure distribution, i.e., chemical ac tion of the slurry chemicals on the material, interplay of the different complexing agents, oxidizers and corrosi on inhibitors and the mechanical abrasion of the particles on the wafer. Chemicals such as oxidizers and corrosion inhibitors vastly affect the reaction rate of the slurry with similar particle nature, size and distributions.
PAGE 23
11 Fig. 1-5. Surface layer formation on Cu with various slurry chemistries. [20] Fig. 1-5 shows the variation of the reaction rate of the different slurry components on Cu when the reaction kinetics were studi ed using electrochemical chronoamperometry [20]. It can be seen from the figure that the surface rate kinetics reach approximately 60 /s when Cu is immersed in distilled wate r. The reaction rate in creases to around 120 /s when 5% H2O2 is added. However upon addition of 10 mM BTA, the reaction rate decreased considerably. Therefore, the chemi cal contribution is very important in the CMP process. The other important factor is the down pressure used. However, for effective CMP of materials, the enhancem ent of removal rate by increasing the down force should not be at the cost of surface quali ty (for example, increase in wafer to wafer and within wafer non-uniformity, delamination, dishing, erosion, etc.) [21, 22] of the wafer. For this reason, the process throughput is compromised to polish the wafers at lower applied load, thereby achieving globa l planarization and be tter surface quality.
PAGE 24
12 Global planarization In contrast to glass or si licon wafer polishing, the objective of the CMP process in interlayer dielectric (ILD) polishing is to planarize the topogra phy created by previous semiconductor processing steps. The pad and slurry play key roles in achieving global planarization, concurrently maintaining the removal rate and producing a defect free wafer surface. If mechanical removal predominates, hi gh frictional forces are produced which result in surface defectivity. Also variation in local polishing pressure leads to variable removal rates within the wafer, which seri ously compromises globa l planarization [20], whereas, if chemical removal dominates, etch ing and corrosion [23] can occur. The key to a good polishing process is dependent on synergy between chemical etching and mechanical planarization. For this purpose, a passivation layer is created on the wafer surface by reaction with slurry a nd removal of this passivation (or modified) layer occurs due to mechanical abrasi on of slurry particles ( Fig. 1-6) For example, in Cu polishing, the formation of the passivation layer is achieved by oxidizers such as hydrogen peroxide, potassium ferricynate, ferric chlo ride, and ferric iodate. Corrosion damage to the surface is prevented by adding corrosion inhi bitors such as benzotriazole (BTA). In tungsten polishing, a rapid formation of surf ace passivation layer is due to the use of peroxygen compounds and stabilizing agents [24] The purpose of the passivation layer in the case of silica polishing is to soften the surface, which is inherently hard resulting in minimal material removal. For this purpose, maintenance of alkaline pH is sufficient in most cases [25, 26]. During Ta polishing, formation of stable Ta2O5 helps in uniform removal of material from the surface [27] To achieve optimal CMP performance, the time to form the passivation layer of desired thickness should be minimized.
PAGE 25
13 Fig. 1-6. Micro and nanoscale phenomena during CMP [20] Selectivity The difference in the removal rate of one material versus another in a given system represents the measure of the slurry selec tivity for these two materials. The higher the slurry selectivity, the more e ffective the end point detection is for the given CMP process. This is due to a marked change between the tribological properties of the material being polished, and the underlayer. The difference in tribological properties can be monitored in-situ using techniques such as motor curren t, force or acoustic emission for better end point detection. Selectivity is a very impor tant criterion in designing any slurry for multilevel metallization. For STI, the slurry must affect the oxide that is being planarized, and not the underlying nitride laye r. For polishing Cu, the slurry needs to react selectively on Cu and spare the barrier la yer Ta and underlying layers of silica or low-k dielectric material. Typically slurry se lectivity of 10 has been repo rted for Cu polishing [20].
PAGE 26
14 The selectivity could be cons iderably improved up to 1000 by introducing a particle free slurry [28, 29]. The reduction in mechanical component due to lack of abrasive implies that the majority of material removal occurs due to the solution ch emistry which can be made highly selective. Addition of specialized chem icals that can act as catalysts in the chemical interaction between slurry and the polished ma terial thereby increasing the rate of the chemical reaction considerably is a co mmon approach for improving the slurry selectivity. Tetramethyl ammoni um hydrate can be added to Cu slurries to considerably decrease the silica polishing rate [30]. An increase in the polishing rate of TaN with addition of phosphoric acid in alumina and colloi dal silica slurry is also observed. Mixed abrasive slurries (MAS) contai ning alumina/silica and alumin a/ceria particles have been developed to improve the sele ctivity of the CMP slurry. Th e details of the experiments with these novel mixed abrasive slurries have been published by Jindal et al. [31, 32]. Though highly selective slurries might help in avoiding defects such as erosion of sub layers, planar defects such as dishing s till exist causing waferto-wafer nonuniformity (WTWNU) due to the differential polishing pressures in the process. Measurement Techniques The outputs of the CMP process are quantif ied by a host of metrics. Material removal rate (e.g., nm/min) is commonly measur ed in terms of thic kness loss per time or weight loss per time. In terms of global planar ization at the wafer scale, film thickness is measured at points across the wafer surf ace and used to calculate within-wafer-nonuniformity (WIWNU). Most commonly, a 49-point measurement grid is used to determine the WIWNU, which is the variation of those measurements with respect to the average thickness. Similarly, a within-die nonuniformity (WIDNU) calculates a similar
PAGE 27
15 variation but just across the area of one die. At the feature-scale of the process, metrics such as step height reduction are used to indicate efficiency of the CMP process. Selectivity of the slurry is more often char acterized by estimating the erosion or polishing of the underneath layer. End Point Detection Theoretically, the end of a polishing step can be realized by setting a time limit to the process. However, this is only a rough estimate because varia tions of polishing pad surface property, fluctuations in slurry chemis try, particle size dist ribution, feeding rate, pad conditioning issues, and other potential variables can cause ove ror under-polishing. Therefore, in-line monitoring or automatic e nd-point detection of CM P is needed. Initial oxide CMP relied on process times and control of removal rate stability to achieve target thicknesses. As the CMP environment is hostile for in-situ metrology, the earlier endpoint techniques relied on indirect methods. One of the commercialized indirect methods monitors the motor current cha nge [33] while driving the carr ier or platen rotation. This technique measures changes in friction forces between the pad and the wafer surface. It is mostly successful for tungsten CMP than for copper or STI CMP processes, because the frictional forces between the polishing pad and exposed films do not differ significantly in the case of copper or STI scenarios. The method sufficiently sensitive for STI, dielectric and metal CMP processes is the vibration monitoring me thod [34]. Vibrations during the CMP process are det ected using an acceleromete r on the carrier behind the wafer. This method is used to detect delami nation of low-k dielectric films in copper dual damascene process [35]. Optical reflectometry is the preferred method for endpoint detection. But, the method is highly intrusive, requiring extens ive polisher modifications. Various other
PAGE 28
16 endpoint techniques have been developed in recent years, many of which are currently adapted to the high volume production stage [35]. CMP Consumables The CMP market in the year 2003 was $1.7B with slurry (24%) and pad (15%) amounting to 40%. The CMP equipment was estimated to comprise 55% with the remaining 6% accounting for conditioner and othe r consumables. The market is expected to grow to $3.3B in the year 2008. The rapi d growth of CMP and its application for polishing various materials, necessitated deve lopment of a variety of slurries, pads, complex process recipes, more complex slur ry mixing and distribution systems. Polishing of different materials and customized n eeds of the various semiconductor industries resulted in more complex CMP equipment with different process dynamics such as linear, orbital and fixed head machines [ 35]. The process of CMP has gone through a lot of evolution from first being used just for s ilicon dioxide planariza tion to the present day planarization applications in pre-metal dielectric (PMD), ILD, STI, metal and gate oxide, etc. While the CMP process seems fundament ally easy to understand, there are a staggering number of process variables (F ig. 1-7) that complicate the underlying interaction between the pad, slurry and wa fer, preventing full comprehension of the process. The most widely accepted mechanis m for the material removal is that the chemicals in the slurry react with the surface of the wafer (e.g., oxide, copper, etc.) and create a modified layer that can be removed easily. The mechanical action of the abrasives trapped in pad asperities then physically abrade and remove the modified material [24, 36, 37].
PAGE 29
17 In terms of the CMP input parameters the process variab les are generally categorized into chemical and mechanical parameters. The chemical nature of the CMP process typically relates to the constituent makeup of the slurry and how those chemicals react with the wafer surface to create a passivated or modified layer, keeping the particles well suspended in the slurry. The chemicals in the slurry are highly dependent upon the type of CMP being used. For example, in oxi de polishing, an alkali ne solution is used along with various buffering and oxidizing agents, whereas for metal CMP as in Cu CMP, acidic slurry can be used. The mechan ical aspects of CMP largely depend on the polishing pad, wafer pattern, and process condi tions like pressure and relative pad-wafer velocity. Fig. 1-7. Parameters involved in CMP process.
PAGE 30
18 Polishing in its simplest sense is controll ed chemo-mechanical material removal to produce a globally flat, defect free surface. The main components of CMP are the CMP slurry and the pad. The complexity of thes e CMP components and their interactions are discussed further in the following sections. CMP Slurry CMP slurry consists of two major co mponents: abrasives and chemicals. Depending on the material of the wafer, the chemistry of the slurry, and the synergy between them, each kind of slurry behaves diffe rently [38]. CMP is influenced to a great extent by numerous slurry parameters such as pH, solution chemistry, particle charge type, concentration and particle size dist ribution of abrasives, complexing agents, oxidizers, buffering agents, surfactants, corros ion inhibitors, etc.[19, 39]. An ideal CMP slurry should be well suspended with good slur ry stability, and be able to achieve high removal rate, excellent global planarizati on, prevent corrosion (in case of metal, especially Cu), and result in good surface fi nish, low defectivity and high selectivity. Rate of surface reaction The reaction between the wafer and slurry chemicals to form a modified layer of desired thickness should be fast enough to avoid material removal by mechanical indentation of the bare wafer. The time scal e in which the passivation layer is formed should be less than the average time of successive particle indentations into the wafer for abrasion to achieve optimal CMP performance. Fig. 1-8 shows the electrochemical chronoamperometry (potentiostatic) analysis of Tungsten by first keeping the samples at cathodic potential to avoid surf ace oxidation and then anodi zing them. The generation of current, which corresponds to surface reac tion rate, is monitored on a millisecond scale
PAGE 31
19 as shown. It can be seen form the figure that Tungsten surface quickly passivates which is conducive for mechanical removal of the material. Fig. 1-8. Chronoamperometry measurements for tungsten [20] For selectivity purposes, the reaction of slurry chemical s with the underneath layer should be significantly low compared to the su rface layer. In terms of operational hazard, the chemicals used in the slurry should not make it hostile for the operators performing CMP and the post-CMP chemicals should be treated to avoid environmental damage. Slurry agglomeration The ideal slurry will have abrasives behavi ng as discrete single particles during the CMP process. However, particles in a real CM P slurry apart from being discrete can also exist in the form of agglomerates [38, 40] An agglomerate is a collection of particles that come together into packed clumps due to phys ical or chemical inte ractions and act as a single entity. Although CMP slurries are compos ed of very fine particles up to 200 nm (0.2 m), large particles of 1 m and greater are often obs erved during the process. Such particles can be due to agglomerati on or contamination from foreign material. These are not desirable for CMP as they can cause micro scratches and pits due to deep
PAGE 32
20 indentation or non-uniformity due to varyi ng polishing pressure. Even minimal and undetectable concentration of larger sized particles are shown to produce surface defects in CMP [41]. The agglomeration of slurry can be prevented by proper slurry design, which includes appropriate use of chemicals, and/or surfactants and the suitable order of their addition to design the slur ry [42]. Other than adding chem icals, the particle surfaces can also be altered to make the suspen sion stable during the CMP process [42, 43]. In order to prevent the particle aggl omeration during the distribution stage, dispense filters are used which filter out the agglomerated particles before slurry delivery. Care is taken so that the filtration of th e agglomerated particles does not change the particle distribution and slurry composition. Th e drop in slurry pressure and flow due to filtration is minimized by using a series of filters instead of a single filter. Chemical Mechanical Polishing Pads The pad traps the slurry particles on its asperities, and executes the polishing action by transmitting the normal and shear forces fo r polishing, thereby playing a very crucial role in process optimization [19]. The pol ymeric polishing pad must have sufficient mechanical integrity and chemical resistance to be effective for numerous polishing runs through its life. The pad must be able to with stand the aggressive sl urry chemistries used in CMP without degrading, delaminating, warpi ng or blistering. Slurry chemistries range from highly alkaline near pH 11 used for po lishing of inter-layer dielectric to highly acidic near pH 2 used for metal polishing. Slur ries also contain oxidi zing agents such as hydrogen peroxide and potassium iodate. The pa d material should be resistant to such environments. As the polishing slurries ar e aqueous based, pad s hould be hydrophilic to allow circulation of slurry to whole wafer during polishing.
PAGE 33
21 The polishing pad is made up of a matrix of cast polyurethane foam with filler material to tailor the hardne ss of the polyurethane impregna ted felts for the specific CMP process. Polyurethanes have a unique prope rty of combining high strength to resist tearing during polishing a nd, an acceptable level of hardness and good abrasion resistance to prevent excessive pad w ear during polishing. The CMP pads can be categorized into four types differentiated by th eir microstructure [44] They are: Polymer impregnated felts (type 1), Poromerics (t ype 2), Filled polymer sheets (type 3), and Unfilled textured polymer sheets (type 4). Pa ds of type 1 and 2 are shown in Figs. 1-9 and 1-10. The figures indicate the variety of pad surface textures or asperities involved in CMP. The polyurethane polymer consists of two major components/blocks of alternating hard organic isocyanates and soft polyols. Each of these segments has different properties, the characteristic properties of a given pad de pend on the relative ratios of these hard and soft segments. Isocyanates, th e more rigid blocks, influence the hardness, modulus, tear strength and stab ility, specifically at elevated temperatures. Polyols on the other hand tend to affect the elasticit y, toughness, abrasion re sistance and the low temperature stability of polymer. For uniform polishing, the pad needs to be in contact with the whole wafer surface with the same contact pressure and relative ve locity between pad and wafer in all contact regions. But wafers are not fl at and typically have some degree of curvature due to stresses induced during manu facture process or differing coefficients of thermal expansion of the various deposited oxide a nd metal layers on the substrate. This necessitates the polishing pad to be flexib le enough to conform to the wafer-scale flatness. This is achieved by laminating a sti ff polishing pad to a fl exible underlying base
PAGE 34
22 pad, which is typically a more compressible material. This improves polishing uniformity across the wafer without unduly compromising th e stiffness of the top polishing pad. A stacked pad consisting of an IC 1000 top la yer on a Suba IV sub-pad (Rodel Inc., Newark, DE) is currently the pad system of choice to obtain both good global wafer uniformity and high removal rate for various CMP systems. (a) (b) Fig. 1-9. Type I CMP pad. (a) Surf ace. (b) Cross-section [44]. (a) (b) Fig. 1-10. Type II CMP pad. (a) Surface. (b) Cross-section [44].
PAGE 35
23 Effect of pad grooves Polishing pads have macrostructure, which can be either perf orations through the pad thickness or surface groove designs. The motivation behind designing pad grooves is to prevent hydroplani ng of the wafer, ensure uniform distribution of slurry over the wafer, and to remove the polishing debris from the pad surface through these chemicals [44]. The groove designs can be circular with either concentric or spiral grooves, crosshatched patterns arranged as an X-Y grid across the pad surface, other regular design such as hexagons, triangles and tire-tread t ype patterns, or irregul ar designs such as fractal patterns, or their combinations. The effect of pad textur e on tribological and kinetic properties of polishing pad has been studied by Philipossian and Olsen [45]. Real time monitoring of coefficien t of friction (COF) was done to estimate the normal shear forces originating during a particular CMP processes fo r a given pad. Frictional and removal rate data were taken on Rodel IC 1000 flat, perforated, XYand K-grooved pads. The Somerfeld number, COF, a new parameter called tribological mechanism indicator (TMI) and materials removal rate were some of the parameters monitored during the experiments. Philipossian and Ol sen [45] reported that flat and XY-grooved pads begin in boundary lubrication and migrat e to partial lubrica tion as Sommerfeld numbers increase. Thus different pad surface te xtures showed different material removal mechanisms. However, the variation of the mechanical properties need to be studied further on a microscale to gage its impact on the CMP process and material removal mechanism. Physical and chemical chan ges in pad during polishing Since polishing is a wet process, the mechan ical properties of the pad must not vary considerably when the pad is wet during CM P. However, the hydrophilic pad material
PAGE 36
24 looses modulus and hardness and gains ductility because water acts as a plasticizer. These changes are reversible [44] and pad properties revert to the dr y values, when the pad is in dry state, indicating that the chemical at tack is minimal [46]. The surface of the polyurethane pad has been studi ed and reported in literature [47] The quantitative analysis of the physical and chemical cha nges that occur in th e polishing pad during polishing was performed by Lu et al. [48]. In thei r research, Lu et al. studied the effect of polishing cycles on the pore size and shape of the commercially available IC 1000/Suba IV polishing pad. It was reported that the por e size and shape distri bution were modified by polishing cycles. Polishing induced a perman ent distortion in the radial direction due to the oscillatory motion of the wafer, while no change was seen in pad pore size in the transverse direction (direction of pad motion). The change wa s attributed to pore closure by surface reflow of the polymer during polishing or conditioning and not due to debris filling. The surface profiles gene rated using white light interferomery showed that the overall micro roughness of the pad decreases with polishing and there is a smoothening effect over a period of time due to polishing cycles. Along with physical changes, the pad undergoes surface chemical modification and degradation due to polishing slurry chemicals. No pad degradation of the bulk ma terial of the polishing pad was observed. The change in spectrum (Fig. 1-11) and shifting of the peaks of the used pad as compared to the new one has been attributed to the realignment of the molecules of the damaged pad surface layer. The porous structure of the polishing pad encourages water seepage. The presence of water in the polymer pore significantly affects the mechanical properties of the polishing pad. Li et al. [47] studied the effect of pad soaking time on the mechanical
PAGE 37
25 properties of the polishing pad using dynami c mechanical analysis (DMA). The dynamic shear modulus decreased to two thirds of its original value when th e pad is soaked for 5 h. Further, the rate of decrease in dynami c shear modulus decreases at around 14 h but does not reach a steady state. No significan t change in the rem oval rate was observed when pads with different soaking time are used in oxide polishing. This shows that removal rate is more of a surface characteristic of the pad and is not affected by the bulk material. Fig. 1-11 The absorbance spectra of used and dry IC 1000 pad along with dried silica slurry. [48] Modification of the pad-wafer contact m echanics can directly impact the CMP process performance. The modified surface can even change the mechanism of material removal during a polishing process [49]. During polishing, the pad surf ace layer degrades decreasing the pad surface roughness. The pad surface which is flattened or glazed needs to be rejuvenated to bring the polishi ng performance back to the original value. The roughening of the pad by a diamond condition er is known to resu lt in pad revival. Along with wafer polishing, the conditioner made of diamond grit considerably alters the surface of the polishing pad. The process of conditioning essentially removes some amount of top pad surface and is performed eith er in-situ (during the polishing run) or ex-
PAGE 38
26 situ in between two polishing runs. The pad damage due to insufficient conditioning is termed as glazing. Experiments were performed on the CMP pad by Lawing [49] at different conditioning levels of aggressiveness (expressed as low, medium and high). The change in polishing rate corre lated better with the predicte d surface contact area rather than the average surface roughness. Effect of temperature change on polishing pad The polishing pad might be subjected to el evated temperatures during the course of the CMP process [19]. The heating is due to the frictional forces generated by solidsolid contact between pad and wafer during polishin g. There is some alle viation of heating when the polishing is done in hydrodynamic c ontact mode [48]. However, there is an overall increase in the pad temperature ir respective of the mech anism of polishing. The effects of pad heating are compounded if the chemical reaction between slurry and pad is exothermic [38]. The mechanical, physical and chemical properties of the polyurethane material can be permanently or temporar ily altered if the pad is heated beyond a particular temperature [50]. DMA, th ermomechanical analysis (TMA) and thermogravimeteric analysis have been used for material characterization of the polishing pad [47, 48]. The applicability for DMA for polishing pads has been elaborately studied by Lu et al. [48]. Based on the results of previous experiments performed by Li et al., [47], DMA was performed to evaluate the structurepr operty relationship of polyurethane pads. The glass transition temperature is the temper ature at which the macromolecules of the polyurethane break away from entanglem ents and undergo micro-Brownian motion. Investigation of glass transition temperature of the polyurethane pa d is important as it
PAGE 39
27 corresponds to a sharp decrease in the hardness and modulus of the pad which in turn affects the removal rate a nd global planarization. Non-uniformity of the polishing pad The pad might be directly responsible for several process defects like WTWNU where there is non-homogeneity of polishing wh en one wafer is compared to another or WIWNU where there is non-homogeneity of pol ishing at different areas of the same wafer. The scanning ultrasound transmissi on (UST) is a nondes tructive technique developed that works on the principle of ultrasound permeability through absorbing visco-elastic medium [51, 52]. The difference in the ultrasound absorption in the areas of varying density and viscoelasticity is used to determine the non-uniformity within a single pad there by giving an in-depth idea of the physical characteristics of the given pad. In order to improve the yield of the CMP process and, to get a highly planar, defect free uniform wafer surface with reduced overall manufacturing costs involved, there is a need to extensively study the fundamental properties of the CMP pads on the whole. CMP process in the future The semiconductor industry has effectiv ely adapted its CMP technology for the 300 mm wafer [53]. Beyond the adoption of coppe r interconnects, several technologies are necessary to continue the sh rinkage of device dimensions and the increase of packing density in ULSI manufacturing. The use of ultr a low-k materials as in terlayer dielectrics has been at the forefront for decreasing the C part of the RC delay. However, polishing of ultra low-k dielectr ics which are mechanically weak (soft) is a daunting task. The single and dual damascene structures comp rising of ultra low-k porous or polymeric materials are more prone to buckling and cr ushing failures due to significantly lower hardness and Youngs modulus as compared to silicon dioxide. Acco rding to the ITRS
PAGE 40
28 roadmap [12] materials with dielectric constant 2.2 will be integrated in the IC by year 2007. Also, there needs to be a marked improvement in slurry selectivity for accurate end point detection when the constituent layers of the damascene structure namely, metal, hard mask, cap layer, barrier layer and di electric are polished. Furthermore, it is necessary to explore the niche of the CMP process in STI and other applications such as backside polishing and the fabrication of micro-electro-mechanical systems (MEMS). Here, the CMP process must be integrated ho rizontally and vertical ly to achieve high throughput and performance. In MEMS app lications, horizontal integration ensures reliability and good performance of a specifi c CMP process run. The development of new polishing pads and pad architectures, novel slurries, new metrology techniques, etc. comes within the scope of horizontal CMP integration. Vertical integration ensures success of every successive CMP operation. This includes the integration of the upstream processes such as Cu/barrier deposition, et ching and downstream processes such as ILD deposition and lithography. This opens a wide scope of research for further optimization and development in Cu CMP process. Due to intricate interactions of the processes involved in any CMP process, modeling e fforts are needed to get a systematic understanding of the process and pr edictability of the CMP outputs. Scope of Dissertation The importance of polishing and the vari ous available planarization techniques with emphasis on CMP have been discussed in Chapter 1. A general overview of the parameters involved in CMP are discussed w ith prominent interactions between them. Chapter 2 gives an overview of the CMP m odeling efforts and the framework of the PERC I CMP model. The models challenges are described, which necessitated the
PAGE 41
29 development of the PERC II model. The assump tions of the PERC I model are refined as necessary in chapter 3. Model development is also delineated with critical differences between PERC I and II models. Chapter 4 va lidates the PERC II model comparing to experimental literature data and comments on the predictive capability of the model. Chapter 5 summarizes the findings of the st udy along with the suggestions for future work.
PAGE 42
30 CHAPTER 2 CMP MODELING: OVERVIEW OF PERC I Like most semiconductor manufacturi ng processes, simultaneous tuning of process variables is a formidable task due to many varying parameters. Unfortunately, much of the commercial CMP process deve lopment cycle is based on nearly blind design-of-experiment (DOE) testing. While these empirical tests help to optimize parameters in a very small range, the overall approach does not necessarily facilitate a deeper understanding of the process. Thus, CMP research has focused on developing process models in order to gain further insi ght on the process mechanics for better control over the CMP outputs. The ultimate goal is to model at three scales of CMP, namely, the particle, die and wafer scales. The models at the particle scale are needed to understand the mechanisms involved in interactions between slurry pa rticles, chemicals, polishing pad and wafer during blanket wafer polishing. The models at the feature or die scale are needed to address the topography evolution of integrated circuit (IC) chips as a function of pattern density, line width, pitc h width and polishing time. The wa fer-scale models also need to address the issues related to the material removal non-uniformity over the entire wafer surface. To satisfy the manufacturability requi rements in the sub-micron IC fabrication, the margin of the profile non-uniformity is us ually limited in the range of several hundred nanometers. The material removal non-uniformity across the wafer is rooted in the nonuniform distributions of pressure and veloci ty over the wafer surface, which in a large
PAGE 43
31 degree, are determined by the configurations of CMP equipment. A mo del at this scale, therefore, needs to consider the e quipment configurati ons appropriately. Literature Review In the last decade, various kinds of m odels have been proposed by researchers, concentrating on different aspects of CMP cove ring the particle, die and wafer scales. But the effectiveness of die and wafer scale mode ls depends strongly on the robustness of the particle scale models, as the die and wa fer-scale models are developed based on the particle scale models. Therefore, in order to understand and predic t the performance of CMP in wafer scale, the current study aims to understand the behavior of CMP at particle scale. As the current work focuses mainly on particle scale modeling, only related models will be reviewed here. Preston Equation The Preston equation [54] as an empirica l relation for optical glass polishing was substantiated by Cook [24] who co nsidered that the interacti on between abrasive particles and wafer surface leads to removal of atomic clusters due to bond breakage. Preston provided a simple model of material remova l in glass polishing tools which is often adopted for CMP processes also. But, experime ntal results have shown that slurry and pad properties also significantly influence the process along with down pressure and platen speed [55, 56]. In or der to explain the non-Prest onian behavior, many modeling efforts have been carried out in the past decade and is still a topic of considerable interest. The three most important elements in the CMP material removal process are the wafer being polished, slurry and polishi ng pad. Researchers adopted two main approaches in modeling the CMP proce ss. One approach is based on fluid hydrodynamics, where material removal is cons idered to be caused mainly by slurry
PAGE 44
32 erosion. Slurry film thickness, hydrodynamic pressure and removal rate are generally calculated by solving Reynolds or Navier-S tokes equation [57-62]. The other approach uses the theory of contact mechanics, which assumes that the material removal occurs by particle indentation into the wafer su rface. Pad surface roughness and hardness are generally used to calculate th e particle indentation into the wafer thereby calculating the removal rate [27, 63-67]. Finite element modeli ng (FEM) has also been used to determine the pad asperity mechanical properties recently [68, 69]. It is generally agreed that the typica l CMP process conditions lead to major material removal by particle abrasion rather than by slurry erosi on [70]. Experimental observations of polished surfaces have revealed that material removal in CMP occurs as a consequence of a combination of chemical reac tion of the slurry chemicals with the wafer surface and ploughing (repeated sliding, rolling, or indentation) of the abrasive particles against the wafer surface that is already m odified. The chemical reaction results in formation of a thin layer on surface whic h is of different modulus than the substrate/wafer, aiding in material removal [71]. When particles are introduced between w earing surfaces, the wear is due to the particles embedded in one of the sliding surf aces. This process is referred to as threebody abrasion. In the CMP process, the particle s acting as single point cutting tools, dig into the wafer and plow a series of groove s in it. Therefore, the CMP process is essentially a three-body abrasive wear. The an alysis of the texture of polished surfaces also supports this observation. Fig. 2-1 illust rates typical post CMP surfaces as examined via an optical surface profiler. The surface texture consists of a number of randomly
PAGE 45
33 oriented nano-grooves (Fig. 21). The width and depth of the grooves are determined by the penetration depth of the abrasive particles into the wafer surface. (a) (b) Fig. 2-1. Polished wafer surfaces showing nanoscr atch surfaces. (a) Si lica wafer [72]. (b) Si wafer [73]. The material removal process in CMP can be considered as sliding of the indented slurry particles over a chemically influen ced thin layer over the wafer surface. The
PAGE 46
34 knowledge of critical factors that effect the CMP proce ss and the interactions among them is critical to select the respective parameters and de sign an enhanced CMP process, predicting material removal rate, and surface quality (roughness, defects, etc.). The most popular material removal equation is the Pr estons Equation [54], which was initially introduced for glass polishing, MRR= KePV, (2 1) where MRR is the material removal rate, Ke an all-purpose coefficient, P the down pressure and V the relative velocity over the wafer-pad inteface. It demonstrates a linear dependency of material removal rate on the pr essure and velocity. Not all experimental MRR data in CMP, especially in metal CM P, support the linear dependency between material removal and the product of pressure and velocity. Modifications to Preston Equation In order to explain the non-Prestonian behavior of the CMP process, revised Prestons equations were proposed by several researchers. For example, considering that the material removal rate does not extrapolat e to zero, Maury et al. [74] introduced a fitting parameter MRR0 into Prestons equation MRR= KePV + MRR0 (2.2) Later, the nonlinear equation MRR= KeP V, (2.3) where and are two fitting parameters, was proposed by Wrschka et al. [75] to get a better fit for the experimental data. The major problem with Prestons equati on and its revisions are that they are empirical modeling efforts. Therefore, they do not shed any light on gaining insight into
PAGE 47
35 the process especially regard ing other process variables. By extending Browns model [76] of the metal polishing to the silic on polishing, Cook [24] developed a physical model to address this limitati on. The interactions between th e abrasive particles and the wafer surface is proposed as a Hertzian elastic penetration [77] of a spherical particle under uniform pressure P into the wafer surface, sliding along the surface with a velocity V and removing material volume proportional to the penetration. The MRR formulation was proposed was MRR= PV/(2E) (2.4) where E is the Youngs modulus of the wafer material. This model can be taken as a theoretical equivalent to the Prestons equation since it suppo rts the linear dependency of MRR on pressure and velocity. The relati onship between the wafer surface roughness Ra the down pressure P, and abrasive size can also be obtained based on this model Ra=3/4 b (P/2k1E)2/3 (2 5) where k1 is the particle concentration, which is unity for a filled close hexagonal packing [24] and b the diameter of th e slurry particles. A similar model was developed by Liu et al [78] based on the statistical method and He rtzian elastic penetra tion. Besides the wafer material properties including wafer hardness Hw and wafer Youngs modulus Ew this model included pad hardness Hp and abrasive Youngs modulus Es resulting in V P E E E E H H H C MRRw s w s p w w e (2.6) where Ce is a coefficient to account for the effects of slurry chemicals and other consumable parameters. This model, similar to Cooks model, suggests that the material removal is proportional to the appl ied pressure and relative speed
PAGE 48
36 The advantages of Cooks and Lius mode ls over Prestons equation are that they provide more insights into the roles of the consumable parameters. The contributions of the slurry abrasives and pad, fo r example, have been attributed to their size and hardness. An additional benefit is that not only material removal rate, but also surface quality issues such as roughness (Eq. 2.5) can be addressed using these models. Slurry Erosion In Cooks and Lius models, the mechanic al removal by abrasive particles was assumed to be the dominant mechanism. However, some researchers instead assumed the material removal to be by slurry erosion approach, i.e., due to mechanical-enhanced erosion. Runnels et al. [79-81] developed an erosion-based model for CMP. They assumed that a fluid film exists between th e wafer and pad interface, which affects the erosion/material removal rates at each si ngle point through the fluid stress tensors. Motivated by observations at feature scal e, modification to Prestons equation was proposed, with relative velocity V, replaced by the tangential stress on the wafer surface; MRR=Cetn (2.7) where Ce is an all purpose coefficient, t is the shear stress due to the slurry flow and n the normal stress. The model uses a steady state incompressi ble Navier-Stokes equation to model the slurry flow at the wafer-polishing pad interface. The wafer surface is assumed to be smooth and spherical with a large radius of curvature. The fluid layer thickness and the angle of attack between the pad and the wa fer are obtained through an iterative procedure satisfying force and momentum balance. Runnels model has been integrated into several particle-scale models by researchers in cluding Tseng and Wang [82] and Zhang and
PAGE 49
37 Busnaina [83]. Tseng and Wang attributed the normal stress at the pa rticle-wafer contact to the elastic indentation of th e particle into the wafer surf ace, which is similar to that proposed by Cook and calculated the normal stress over the wafer-particle interface as 2 c nr F (2.8) where F is the force acting on the spherical part icles, which is proporti onal to the pressure P. The radius of particle-wafer contact, rc is given by E E F d rc 2 2 31 1 8 3 (2.9) where d is the diameter of particles, and the Poissons ratios of wafer surface and the particle and E and E the elastic modulii of the wafer and particles, respectively. The shear stress due to the slurry flow can be approximated as 0VPA Ce t (2.10) where is the dynamic viscosity of the slurry and A0 the area of wafer surface respectively. Substitution of equations (2 8) and (2 10) into (2 7) yields MRR=KeP5/6V1/2 (2 11) where Ke is the parameter to account for ma terial properties, slurry abrasive concentration and chemical processes. This model demonstrated a nonlinear relationship between the material removal and the product of pressure and velocity. Tsengs model attempted to connect the elastic indentation to the erosion rate, whereas Cooks and Lius models connect it to the mechanical ab rasion. While the down pressure dependency (P5/6) is quite close to linear depe ndency, the velocity dependency (V) however, is quite nonlinear. This is because the contribution of velocity has been attributed to the slurry
PAGE 50
38 flow instead of abrasive sliding. Due to pad porosity and surface asperities, such a slurry film is bound to be interrupted by solid-s olid contact. Thus hydrodynamic lubrication may be responsible for distributing the slurr y, while the material removal occurs mainly through solid-solid contact. Cooks, Lius and Tseng and Wangs models attributed the penetration of the abrasive particles to Hertzian elastic contact. Zhang and Busnaina [83] estimated the contact pressure between the particle and the contact surface in their model and found that it is larger than the yi eld stress of the polished materi als. Therefore, they proposed that plastic deformation is a more likely deformation mechanism in polishing. The contact pressure over the particle-wafer interfaces is sugge sted to be equal to the hardness Hw of the wafer materials. Replacing the nor mal stress (Eq 2 8) in Tseng and Wangs model with the hardness Hw yields the following material removal rate formulation MRR= Ke(PV)1/2 (2 .12) where Ke is the parameter to account for ma terials properties, slurry abrasive concentration and chemical processes. Both Tseng and Wang; and Zhang and Busnamas models suggest a nonlinear pressure times velo city dependency of material removal rate. Moreover, it is noted that besides the external force applied on the particles from the pad, Zhang and Busnaina also proposed that an a dhesion force, either va n der Waals force or electrostatic force depending on the separa tion distance between the particle and the wafer, contributes to the indent ation. This has been integrated into another particle-scale model by Zhang et al. [84]. In a series of articles by Ahmadi and Xia [85], and Mazaheri and Ahmadi [86, 87], a thermodynamic work parameter Wa of adhesion was added to account for its effects on the indentation of ab rasive particles. Mazaheri and Ahmadi [86]
PAGE 51
39 introduced a double layer (dl) force into the indentation force, whose magnitude is a function of the zeta pote ntial of the abrasives and the wa fer. Beside the abrasive wear, Ahmadi and Xia [85] also considered the a dhesion wear of the wafer in their model. Moreover, while most of the models treated the abrasives as having a spherical shape, Mazaheri and Ahmadi represented them as spheres with a number of hemispherical bumps around their surface. The penetrations of slurry abrasives are modeled using the penetrations of the bumps. The above models imply that the abrasives are embedded in the pad and indent into the wafer surface. It should be noted that some modeling efforts also assume that the abrasive particles float in the slurry and impact the wafer surface from time to time causing material removal [88]. Pad Considerations in CMP Models The pad is assumed to be smooth in the earl ier particle-wafer interaction models. It has been observed that the pad topography and pad material play an important role in material removal processes. For example, the material removal rate has been reported to increase with the pad surface roughness [55] Without conditioning of the polishing pad surface, the material removal rate decreases exponentially with polishing time [89]. In consideration of this, Yu et al. [63] developed a pad-based model. They approximated the peaks on the pad surface by hemisphe rical asperities with consta nt radius and the asperity heights to follow Gaussian distribution. Base d on the model, the real contact area is smaller than the nominal (apparent) contact area and linearly proportional to the down pressure. Steigerwald et al. [19] proposed that the material re moval rate is proportional to the number of abrasive partic les over the contact area. Combining this argument with
PAGE 52
40 Yus model yields a linear dependency of the material removal rate on the down pressure, which agrees with Prestons equation. Zhao and Shi [90] also proposed a model based on the wafer-asperity contact. Unlike Yus model [63], this model does not consider the Gaussian distribution of the asperity heights. The contact area between an asperity and the wafer is given by Aa P2/3 based on Hertz elastic contact theory. By comb ining Steigerwalds argument, the material removal rate formulation can be obtained as MRR= K(V)P(2/3) (2.13) where K(V) is a function of the relative ve locity V and other CM P parameters. It is further considered by Zhao and Shi [90] that when the particles are rolling against the wafer surface, their contribution to material re moval will be negligible. They argued that whether the particle is rolling or not is determined by the surface friction between the particles and the wafer, and only when the dow n pressure P is larger than a threshold down pressure Pth, the pure rol1ing can be avoided. Th e material removal rate is given as: MRR = K(V)(P2/3 Pth 2/3) for P Pth (2.14a) and MRR = 0 for P< Pth (2.14b) The fundamental difference between the a bove pad-based models and the particlebased models by Cook and others is that pad-based models a ttribute the material removal rate to the number of abrasive particle s captured by the polishi ng pad while the later attributes the material removal rate to the interaction between a si ngle abrasive and the wafer.
PAGE 53
41 A particle-pad interaction is critical to th e particle scale models considering that the function of the pad is to trap the abrasive par ticles and transmit load to the particle-wafer interface. Cooks model [24]suggested a clos e-packing of spherical abrasives onto the pad. It is assumed that the wafer and pad ar e separated completely by the abrasives and no direct pad-wafer contact exists. The for ce applied on a single abrasive under these assumptions is given by 1 2Pr 3 2 k F (2 15) where P is the polishing pressure, r the abrasive size and k1 the particle fill fraction on the pad. This particle-pad interaction model has been integrated into the material removal model of Cook. It is also used by Ahmadi and Xia [85] to evaluate the force on a particle in their case of a hard-pad and a larger con centration of abrasive particles. Later, Zhao and Shi [90] proposed that when the pad is soft enough, the abrasive particles will be embedded deeply into the pad and the force from the wafer is supported by the pad and abrasives together. This idea has been applied by Luo and Dornfeld [27, 91, 92] and Fu et al. [93] in their model. Luo and Dornfelds m odel [27, 91, 92] suggested that this force is proportional to the contact pressure multiplied by the abrasive size by assuming that the abrasives are closely packed to each othe r and these closely packed abrasives are enclosed by the pads so that the effective co ntact area between wafer and pad is equal to that without abrasives. More over, the size of the abrasive s that may be captured by the pad is assumed to be a function of abrasive size distribution and pad properties. Their model, however, neglects the lower sizes of the particle size distribution in material removal calculation. Fu et al. [93] later assu med that the abrasives are dispersed evenly over the pad surface and used a beam model to evaluate the wafer-pad direct contact
PAGE 54
42 between two single abrasives. The force s upported by a single abrasi ve can be obtained from the beam model and is a function of ab rasive size, down pressure and pad material properties [93]. Besides the force, the sec ond purpose of an abrasive-pad interaction model is to evaluate the number of abrasive s involved in the material removal process. Fu et al. [93] considered the number of trapped particles as an independent parameter in their model. This may be misleading considering that vari ous parameters, for example, the abrasive weight concentration and abrasive size, have an influence on the number [27, 72, 91, 92]. Luo and Dornleld considered more complex scenarios and suggested that only a portion of abrasives are involved in the material removal process, which is not reasonable. Slurry Considerations in CMP Models The contribution of slurry chemicals to the material removal is either neglected in earlier models or represented by an all-pu rpose coefficient. Cook [24] suggested a complete but complicated scenario of the chemical effects. Besides the mechanical removal, it is proposed that the surface re moval during polishing include the following five chemical processes (i) the slurry chemi cal diffusion into the wafer surface, (ii) the subsequent wafer material dissolution under th e load imposed by the abrasive particles, (iii) the adsorption of the dissolution product on to the surface of the polishing grain, (iv) the re-deposition of the polishing materials back onto the wafer surface, and (v) the corrosion rate between particle impacts. Thes e steps are not considered in most of the early models as they are complex to model quantitatively. The major contribution of slurry chemi cals on the material removal process has been attributed to the formation of a surf ace layer. This idea is well demonstrated by Kaufman et al. in their Tungsten CMP model [37].
PAGE 55
43 Kaufman et al. proposed that this passivat ion layer is removed by slurry abrasives and the fresh tungsten surface is exposed, whic h is subsequently passivated and removed. This mechanism of continued passivation and removal can be used to explain the copper, aluminum and other metal CMP systems as well. A similar mechanism is supposed to work in silicon, silica and low-k CMP system s. Paul [94] and Zhao et al. [95] have proposed detailed surface kinetics models to connect the slurry chem ical concentrations and fresh metal surface sites available to the formation rate of the surface layer. Their models can explain the material removal rate as a function of chemical and abrasive weight concentration. Borst et al. [62] have proposed a five-step model for CMP of SiLK dielectrics (i) mass transport of reactant from the slurry to the slurry /wafer interface, (ii) adsorption of reactant to available surface site (iii) reaction between adsorbed reactant and specific wafer surface to form an alte red wafer surface layers, (iv). mechanical removal of the altered wafer surface layer, a nd (v) mass transport of removed material to the bulk slurry. In their work, the mass trans portation, slurry chemical concentration and reaction rate to the formation of the surface layer were inco rporated. This detailed model was suggested to be able to be extended to the CMP of other materials. The idea of continuous surface modification and removal has been applied in the model developed by Qin and co-workers [96]. A similar approach has also been used in Luo and Dornfelds model [91]. They exte nded Kaufmans model by proposing that the surface layer is a bi-layer structure, though in reality there might be a gradual transition between the layers. They did not cover the details on how the formation rate of the surface layer is affected by the slurry chemi cals. But Luos model assumes that all the
PAGE 56
44 asperities contact the wafer and that the part icles fill the asperity-wafer contact spots, which is not necessarily true. The slurry chemicals affect not only the wafer but also the slurry abrasives and polishing pad. Their contribution to material removal is therefore not only reflected through surface kinetics, but also through alte ration of the abrasive and pad properties such as the abrasive shape, abrasive si ze and pad Youngs modulus. There are not many modeling efforts on these two interactions ye t. A recent model by Castillo-Mejia et al. [97] tried to explain the effect of water-pad interaction in CMP. The water is proposed to plasticize the polishing pad and reduce its elas tic modulus. A formulation on the ratio of the Youngs modulus of wet pad to the dry pad is suggested. This Youngs modulus is then used in a wafer-pad contact mode l to evaluate the material removal. Many of these models were also extended to patterned wafers along with other modeling concepts including combination of the solid-solid contact and slurry erosion approaches [98-109]. Polishing mechanisms we re also based on surface reaction kinetics [95, 110, 111] and modeled using mo lecular dynamics simulations [112-115]. PERC I CMP Model Most of the earlier models have concentr ated on one or two interactions of the CMP process parameters. They are useful fo r identifying the input parameters. However, they may not be sufficient for understanding the whole material removal process. The effects of the same input parameters may be contrary to each other when acting in different conditions. A comprehensive model of CMP integrating the main interactions together is therefore needed. For this purpos e, a model was develope d by Qin et al.[72], taking into consideration most critical CMP parameters. This section discusses the
PAGE 57
45 assumptions on which the model is built, its effectiveness in predicting the CMP process variables and the model limitations. Model Development The chemicals in a slurry alter the propertie s of the wafer surface that is in contact with them. Since the chemical influence is restricted to a thin region near the wafer surface, a thin surface layer w hose mechanical properties are different from those of the bulk wafer is assumed. This surface layer can be formed by chemical reactions at the surface or by the diffusion of slurry ionic speci es into the wafer. The surface layer is then removed by mechanical action of the slurry part icles and/or by dissolution into the slurry through a convective diffusion process. Alt hough there may be a few exceptional cases [Luo98], material removal by the di ssolution is small compared to that by the mechanical action of the slurry particles, and it is not c onsidered in this model. Fig. 2-2 describes schematically the physical situation that is considered in the present analysis. The pad asperities that are in contact with the wa fer surface act as the anchoring sites for the slurry particles. The anchor ed particles then gouge the ch emically altered wafer surface or the bulk wafer depending on th e indentation depth re sulting in material removal. In the following section, various parameters cons idered in the present analysis and the underlying assumptions are described followed by the details of the model development. Topology of a pad surface has been charac terized and described statistically by various researchers [63, 116] Their measurem ents indicated that pad asperity heights were in the order of tens of microns and appeared to have a normal (or Gaussian) distribution. In Fig. 2-2, the topology of the pad surface is shown schematically along with a statistical description characterizi ng the pad asperities. In this figure, is the
PAGE 58
46 mean radius of curvature of the tip of the pad asperities, is the distribu tion function for the height of the asperities, and the standard deviation of the pad asperity heights, respectively. P V Wafer Pad Asperity Mean Height z (z) h Asperity Height, z (a) (b) Fig. 2-2 Pad consideration in PERC I. (a) Sche matic of the pad surface that is in contact with the chemically modified laye r formed on the wafer surface; (b) Distribution of the pad aspe rity heights (When the height of an asperity is greater than the average gap h between th e pad and the wafer, that asperity is in contact with the wafer surface.) [72]
PAGE 59
47 Since the characteristic length scale fo r the roughness of the pad surface (e.g., ) is much larger than the feature sizes on the wa fer, it was assumed that the wafer surface is smooth and flat compared to the pad surface. This assumption makes it possible to use the Greenwood-Williamson model [117] th at describes the contact area, Areal, between a flat surface and a randomly rough surface under a normal force (or load), L : dz z h z NA Ah o real (2.16) dz z h z E NA Lh o 2 3 2 1 *3 4 (2.17) where N is the number density of the pad asperities (i.e., number of asperities per unit area), Ao is the total area of the flat wafer su rface, and h is the average gap between the smooth wafer and the rough pad under the give n conditions. Then, the down-pressure, P, applied to the wafer with an area Ao is L/Ao whereas the actual mean contact pressure, Pr, is L/Areal if the normal force applied to the wafer is entirely supported by the pad asperities in contact with the wafer. That is L = PAo = PrAreal (2.18) E* is the composite modulus between the two surfaces in contact, that is defined as w w p pE E E2 2 *1 1 1 (2.19) where E and are the Youngs modulus and the Po issons ratio, and the subscripts p and w represent the pad and the wafer, respectively. Equations 2.16, 2.17 and 2.18 provide an explicit relationship between the contact area ( Ar) and the down pressure ( P ) that is needed in modeling the CMP process. From Eqs. 2.16 and 2.17,
PAGE 60
48 dz z h z dz z h z E L Ah h real 2 / 3 2 / 14 3 (2.20) Following Yu et al.[63] the normal distribution function may be used for the distribution of the asperity height, When 2 22 z exp 2 1 zis substituted into Equation 2.20, the integral cannot be obtaine d in an explicit form. Instead, a numerical integration should be preformed. Johnson showed that the ratio of the two integrals in Eq. 2.20 is nearly a constant for a broad range of h / [77]. Thus, 2 / 1 1 2 / 1 1E PA C E L C Ao real (2.21) where C is a constant whose value is between 0.3 and 0.4 when h / is in the range of 0.5 and 3.0. Recalling that is the standard devi ation of the asperity height distribution, h / =3.0 represents a situation where only 0.13% of the pad as perities are in contact with the wafer whereas about 30.85% is in contact when h / =0.5. Practical situation in a typical CMP process is well within this range of h / value. From Eqs. 2.18 and 2.21, 2 / 1E C Pr (2.22) It should be noted that the contact area ( Ar) increases linearly with the downpressure whereas the contact pressure ( Pr) remains constant due to the linear increase of Ar with P The underlying assumption for Eq. 2.18 is that the applied normal force on the wafer is supported entirely by the pad asperities. As is mentioned previously, experimental measurement on the friction co efficient indicates that the hydrodynamic
PAGE 61
49 pressure is negligible under typical CMP operational conditions. Furthermore, commercial pads often have grooves or perf orations to prevent hydroplaning (i.e., to prevent a large pressure development due to a lubrication flow) a nd to ensure uniform slurry distribution. Thus, the assumption that the hydrodynamic pressure is negligible is justifiable. Small Plastic Deformation of the Thin Layer on the Wafer Surface Considering the fact that the deformation of the surface layer by the slurry particles under the contact pressure is ve ry small in the order of seve ral nanometers, the mode of deformation can be assumed to be plastic de formation. In the case of oxide CMP, it is known that the silica surface is softened when it is in contact with an aqueous solution. A surface layer (or so-called gel layer) is form ed due to the diffusion of water molecules into the silica surface alteri ng the surface properties. The situation may be different in the case of metal CMP. Since an oxidizer is added in most metal CMP processes, metal oxide is formed on the wafer surface creating a structure mismatch at the interface betw een the oxide layer and the bulk wafer underneath. Thus, delamination or microcrack ing of the oxide layer may occur at the interface during the metal CMP process. However, as evidenced in previous sections, it is likely that plastic deformati on is still the dominant mode of deformation because the indentation depth by the slurry particles is small (e.g., less than several nanometers). Furthermore, the oxide layer may be soft due to hydration in an aqueous environment. Thus, the assumption that the chemically alte red thin surface layer on the wafer exhibits plastic deformation is plausible. This assump tion enables us to apply the abrasive wear theory in deriving the expre ssion for the material removal rate. When the indentation depth is beyond the surface layer, the deformation of the unaltered wafer material may be
PAGE 62
50 either plastic or elastic depending on the hard ness of the wafer material and the slurry particles. Material Removal by a Single Particle In the present analysis, it is assumed that the slurry particles are monosized and of radius R. The overall material removal rate is assumed to be the multiple of the removal rate by a single particle and the number of particles in contact with the wafer: (RR) = n (RR)v/Ao (2.23) where n is the number of slurry particles anc hored in the region of the pad that is in contact with the wafer, and (RR)v is the volumetric removal rate by a single particle. The overall removal rate, (RR), is typically desc ribed in terms of the thickness removed per unit time; hence the division by the nominal wafer area Ao. Schematic illustration of single particle abra sion is given in Fig. 2-3. The material removal is viewed as a sliding-indentation process in which the volumetric removal rate by a particle is given as (RR)v = AiV (2.24) where Ai is the cross-sectional area of the part icle immersed in th e wafer under pressure (hashed area in Fig. 2-3a), and V is the relative velocity of the pad to the wafer. Assuming that the hard particle maintains its spherical shape, R a R R a Aiarcsin2 (2.25) where is the indentation depth and a the radius of the circular indentation on the wafer surface by the particle of radius R. Because is very small compared to R, a/R is also very small and Equation 2.25 can be linearized about a/R to give
PAGE 63
51 R O a Ai1 3 4 (2.26) Thus the volumetric removal rate is V a RRv 3 4 ) ( (2.27) (a) (b) F a F a Padt Wafer t R b Pad Wafer a Fig. 2-3. Schematic of single particle ab rasion showing the removal of chemically modified surface layer. (a). t. (b). > t with soft wafer materials. Because a2=(2R 2) and is much smaller than R, Equation 2.27 can be rewritten as R V a RRv3 2 ) (3 (2.28) Once the indentation depth (or a) is known, the volumetric removal rate by a single particle can be readily calculated from Eq. 2.28. Depending on the relative magnitude of the indentation depth ( ) to the surface layer thickness (t), the material removal may be classified into two different regimes as depicted in Fig. 2-3; (a) t and (b) > t.
PAGE 64
52 Indentation depth smaller than th e surface layer thickness (i.e., t) When the chemically modified surface la yer is thick enough, the indentation depth will be no greater than the surface layer thickness and only plastic deformation within the surface layer occurs. In this case, (or a ) can be obtained from the force balance imposed on the particle by the pad and the wafe r that depends on the indentation depth of the particle into the wafer su rface. The force imposed on a particle by the pad is given as r aP b F2 (2.29) where b is the radius of the circular indentation into the pad as described in Fig. 2-3a, and Pr is the contact pressure given by Eq. 2.22. The force on the particle by the deforming wafer can be given as e aB a F2 (2.30) where Be is the effective Brinell hardness of the chemically modified surface layer. The term effective is used because its value is influenced by the thickness of the chemically modified surface layer and the material prope rty of the underlying substrate. The pad is supposed to hold the particles in place th ereby transmitting the normal force to the particles that cause indentat ion of the wafer surface. The pad should be soft enough to provide stable anchoring of the particles on its surface. However, if it is too soft, the particles will be embedded too deeply into the pad diminishing the material removal capability of the particles. Thus, the pad shoul d be sufficiently hard to make the value of b in Fig. 2-3 to be close to the particle radius R. Then, from equations 2.29 and 2.30 with b=R,
PAGE 65
53 2 / 1 e rB P R a (2.31) This equation along with equations 2.22 a nd 2.28 gives the following expression for the volumetric removal rate by a single particle of radius R. V B E R C RRe v 2 3 4 3 2 2 33 2 ) ( (2.32) Indentation depth greater than the surface layer thickness (i.e., > t) When the bulk wafer material beyond the surface layer is relatively soft (e.g., copper) compared to the hardness of the particle s, it is possible for the particle to cause deformation into the wafer beyond the surface layer (Fig. 2-3b). As it was described in the previous section, the deformation is expe cted to be plastic for both the surface layer and the bulk wafer, and the force causi ng the deformation may be given as w e aB tR a tRB F) 2 ( 22 (2.33) where Bw and Be are the Brinell hardness of the bulk wafer material and the chemically modified surface layer, respectively. Because this force is equivalent to the one imposed on the particle by the down-pressure, Eq 2.33 is equated with Eq. 2.29 with b=R to give 2 / 1 2) ( 2 w e w rB Rt B B P R a (2.34) Then, Eq. 2.28 for the volumetric removal ra te by a single particle is reduced to R V Rt B B R B E C RRw e w v 2 3 2 2 11 2 3 2 ) ( (2.35)
PAGE 66
54 While equation 2.35 describes a situation where the deformation reaches into the bulk wafer beyond the surface layer, deformatio n of the bulk wafer may not occur if the bulk wafer material is very hard. This lim iting case corresponds to the situation of infinitely large Bw in equation 2.35 that is reduced to V t R RRv 2 / 3 2 / 13 2 4 ) ( (2.36) Equations 2.32, 2.35 and 2.36 indicate that the volumetric removal rate by a single slurry particle depends on the geometry of the pad and the particle (i.e., and R), the material properties (E*, Bw and Be), the thickness of the surface layer (t), and the relative velocity of the pad (thus the particle) to the wafer (V). It may be interesting to note that (RR)v is independent of the down pressure, P. It is due to the fact that the contact pressure, Pr, is independent of P. As equation 2.21 indicates, the contact area of the pad with the wafer (Ar) increases linearly with P. Consequently, the mean contact pressure, Pr, which is PAo/Ar, is independent of P. In the following section, the number of particles in contact with the wafer surface is estimat ed to determine the overall removal rate. Estimation of the Overall Removal Rate When the wafer is pressed against the pad and is in sliding motion relative to the pad, the particles can be entrapped onl y in the region where the local distance between the wafer and the pad asperity is smaller than the particle diameter, 2R. Thus, the number of particles that can be trappe d on the pad asperities is n = 2RAreal n (2.37)
PAGE 67
55 Here Ar is the real contact area between the wafer surface and the pad asperities as it was described in previous section. n is the number density of the particles in the slurry (i.e., number of particles per unit slurry volum e), that can be calculated from the solid loading Ca (wt%) of the slurry particles as a s a nR C 34 3 (2.38) where a and s are the densities of the particle a nd the slurry solution, respectively. Substitute of Eqs. 2.38 into 2.37 gives r a s aA R C n 22 3 (2.39) By combining Eqs. 2.39 and 2.18, the number of particles anchored on the pad asperities becomes 2 / 1 22 3 E PA R C C no a s a (2.40) Now that (RR)v and n are given in terms of various pr ocessing variables and material properties, the overall material removal rate in a CMP process can be estimated from Eqs. 2.32, 5.27 and 2.40 for the two different cases described above: Removal Rate: Indentation depth smaller than su rface layer thickness (i.e., t), PV B E C C RRe a s a 2 / 3 2 / 1 4 1) ( (2.41) Removal Rate: Indentati on depth larger than surface layer thickness (i.e., > t),
PAGE 68
56 PV R t B B B E C E C C RR/ w e w * a s a 2 3 2 1 2 11 2 1 1 (2.42) As Eq. 2.23 indicates, th e overall removal rate, (RR), is the thickness removal rate which is the volumetric removal rate divided by the nominal wafer area Ao. We may recall that C is a constant whose value is about 0.35 when h/ is between 0.5 and 3.0 [Joh85]. The overall removal rate given above is reminiscen t of Prestons equation in that the material removal rate is proportiona l to the down-pressure, P, and the relative velocity, V. It should be noted that the Prestons coefficient Kp, whose value relies on empiricism, is now given as a function of various proce ssing variables and material properties. The model predicts the removal rate as a f unction of pressure and velocity product (PV) very well [72]. The removal rate is usually only considered as a function of PV in many modeling efforts, ignoring other processing parameters. But the PERC I model considers the particle size effect, which is crucial for any CMP system. Equation 2.41 indicates that th e overall removal rate is inde pendent of the particle size R whereas Eq. 2.42 indicates a decrease in (RR) with increasing particle size. When the indentation-depth is smaller than the surface layer thickness (i.e., t), the volumetric removal rate by a single particle (RR)v is proportional to R2 (Eq. 2.32). The number of particles in contact, on the other hand, is inversely proportional to R2 as the Eq. 2.40 indicates. Consequently, the par ticle size effect is negated. Thus the model predicts as shown in Fig. 2-4 that the removal rate is either independent of particle size in the modified layer removal regime or decreases with particle size in the substrate indentation re gime. But, the reported literature has shown
PAGE 69
57 that removal rate also increases with particle size in a given range, and that there is an optimal particle size at which enhanced polis hing can take place as shown in Figs. 2-5 and 2-6. To overcome these limitations, a (PERC II) model is developed by analyzing the crucial assumptions in PERC I, and refining them when necessary. The following chapter deals with assumption validation and PERC II model development. Particle Size Removal Rate Wafer removal Modified layer removal Particle Size Removal Rate Wafer removal Modified layer removal Fig. 2-4. Schematic of PERC I predic tions for increasing particle size. 0 50 100 150 200 01020304050607080 Mean Particle Size (nm)Removal Rate (nm/min) Fig. 2-5. Polishing of silica by nanoslurry showing optimal particle size [118].
PAGE 70
58 0 100 200 300 400 500 600 256095130 Mean Particle Size (nm)Removal Rate (nm/min) 18 kPa, 0.4 m/s 32 kPa, 0.6 m/s 45 kPa, 0.9 m/s Fig. 2-6. Polishing of silica at various pr essure, velocity and pa rticle sizes showing optimal particle size [116].
PAGE 71
59 CHAPTER 3 PERC II CMP MODEL The previous chapter highlighted the basic concepts involved in PERC I model [96] development, along with various challenges faced by it. In this chapter, some of the critical model assumptions are verified a nd refined as necessary. The changes are incorporated into the basic framework of the PERC I model to develop the PERC II CMP model. Assumption Validation The critical components in the CMP proce ss are the wafer, pad and slurry. As the wafer is fixed in a given CMP process, th e contributions from pad and slurry are investigated to include their most crucial char acteristics into the model. The main role of the pad in CMP is to trap the abrasive partic les for material removal and transmit the load applied onto the trapped partic les. In particle scale modeli ng, the function of the pad in global planarization need not be considered. Pad-wafer Contact Area The pad-wafer contact area strongly influen ces the number of part icles that can be trapped under each pad asperity. The greater the contact area between pad and wafer, the higher the number of particles th at can be trapped resulting in enhanced material removal. The changes in pad properties during polishing are important to model the CMP process. Often dry properties of the pad are investigated [119], rather than the more appropriate wet properties. The pad charac terization investigations are also limited to the bulk [48,
PAGE 72
60 68, 120] of the material rather than the su rface. These bulk properties in dry state are usually employed in conventional CMP modelin g efforts [68]. Since, polishing involves the interfacial behavior of pa d, abrasive, and wafer, pad surface properties should be used rather than the bulk properties. Surface pr operties of the pad rapidly change due to immersion in slurry [121, 122], quickly reaching equilibrium values [46]. There is one fundamental question that n eeds to be answered. What is the real contact area between pad and wafer? A nd is the Greenwood-Williamson model [117], which states that contact area is linearly pr oportional to the applied load applicable in CMP systems? A basic assumption of the PERC I model is that removal rate is linearly dependant on the pad-wafer contact area. Abrasive particles trapped between pad asperities and wafer are considered to be re sponsible for material removal. As applied load on the wafer increases, the pressure at the pad asperity-wafer interface is supposed to remain constant due to elasticity of the pad, while the pad-wafer contact area increases. This behavior is predicted by the Green wood-Williamson model, which assumes that the pad asperity tips contact the wafer under Hertz elastic theory conditions. Although a number of CMP models have employed this concept [63, 109], no systematic attempts are reported that provide validity of this assumption. Similarly, some researchers have assumed the contact area to be proportional to the power of two-thirds [90] of pressure without any experimental basi s. In the past, pad-wafer c ontact area has been generally derived mathematically [27, 123]based on sta tistical surface roughness rather than actual experimental measurements. The pad-wafer c ontact area reported in literature including theoretical calculations varies from 0.0005% to 20% [27, 49, 63, 109, 124, 125]. Considering the importance of reliable valu e of the pad-wafer contact area and its
PAGE 73
61 dependence on applied load in CMP modeli ng, a systematic attempt was made to measure it under CMP conditions. Kim et al. [ 124] have measured the pad-wafer contact area using optical microscopy and have reported a value of 5%. On the other hand, Basim [125] has reported the value to be 0.3% in dry state based on FTIR measurements. To investigate the pad-wafer contact area, a relia ble technique is needed. To establish the reliability of optical microscopy and/or FT IR techniques, contact area measurements were performed on pad using optical microscopy. Optical microscopy The average pad asperity radius is around 30 m for a typical CMP pad and the RMS roughness around 5m. Therefore, an optical micros cope can be used to study the contact spots, the majority of which can be expected to be over a micron in size. To simulate the wafer, a microscope coverslip was used. To minimize the bending of the coverslip under applied load, a fixture was fabr icated as shown in Fig. 3-1 which enabled easy access to the microscope objective lens for translation in in-plane direction. The fixture consists of a base plate with a window hole of 10mm x 5mm to allow optical observation of the contact spots with mini mum bending of the coverslip under applied load. Optical measurements were performed in dry environment using Leika microscope equipped with frame grabber. One inch s quare coupons were punched out from IC1000 pad, which was used for polishing. The pad was ex-situ conditioned using a grid-abrade diamond pad conditioner. Minimal contrast be tween contact and non-contact regions was obtained, when the pad specimens were analyz ed under the given load. This was not even sufficient for qualitative anal ysis. Measurable distinction between contact and non-
PAGE 74
62 contact spots was achieved with coating of the pad surface with a 5-10 nm thick goldpalladium layer using the Denton sputter coater. The thin coated layer is not expected to affect the pad-wafer contact area results, as the attempt here is to measure the contact area under dry conditions. Fig. 3-1. Schematic of the fixture used in measuring pad-wafer (dry) contact area. Measurements were performed in two mode s. Mode I consisted of measuring the contact area of selected aspe rities as a function of applied load and mode II consisted of measuring various contact spots under appl ied load by translating the microscope objective to cover the entire sp ecimen in view. The optical micrographs were analyzed using Spot and Image-Pro softwares to calculate the percent contact area. The micrographs obtained in mode I are s hown in Fig. 3-2. The regions P and N indicate the pad asperities that are of interest in all three figures, though only Fig.32a is labeled for clarity. In P regions, the asperity contact increases with applied load, whereas, in N regions, new asperity contac ts are formed with applied load. The padwafer contact area values obtained from optical measurements are plotted in Fig. 3-3. The
PAGE 75
63 results indicate that the contact area of a given pad asperity increases linearly with applied load and more importantly the magnit ude of contact area obtained from optical technique is in the same range as obtained from the FTIR technique by Basim (Fig. 3-4). The discrepancy between this data and Kim s optical microscopy result [124] could be due to the possibility that some non-contact areas were counted as contact spots due to poor contrast, as no pad surface modification or coating was mentioned in their study. N P N P (a) (b) (c) Fig. 3-2. Pad asperity contact spots seen through optical mi croscope. (a) 0.75 psi. (b) 5 psi. (c) 8.8 psi. 0 0.2 0.4 0.6 0.8 1012345678 Applied pressure (psi)Pad asperity contact area (%) Fig. 3-3. Pad asperity contact area dependen ce on applied load. Each legend represents a selected asperity contact. Various as p erit y contacts at same p ressure
PAGE 76
64 Fig. 3-4. Pad contact area as a function of applied load [125]. The main motivation behind investigating contact area measurements is to get a realistic value of pad-wafer contact area unde r CMP conditions. The optical and FTIR techniques are used in dry conditions, whereas the actual CMP is a wet system. Combining optical data from this study and FT IR data from Basim [ 125], it is established that FTIR with attenuated total reflection (FTIR-ATR) method can be successfully used to estimate the pad-wafer contact area. The same technique was employed in the present investigation to measure the contact area in wet condition simulating the CMP environment. Fourier transform infrared spectroscopy (FTIR) To investigate the pad-wafer contact area variation with applied load in various slurries, FTIR-ATR was used. The contact spot s contribute toward the absorbance signal resulting in quantitative estimation of th e pad-wafer contact area depending on signal
PAGE 77
65 intensity. Maximum signal intensity is obtai ned when the pad surf ace is in full contact with the wafer surface. The contact area is obtained by normalizing the measured signal intensity with the maximum signal intensit y [126]. The setup used for contact area measurements is shown in Fig. 3-5. IR Beam Pad Applied load ATR crystal IR Beam Pad Applied load ATR crystal Fig. 3-5 Schematic of the FTIR-ATR technique used to measure pad-wafer contact area. A silicon ATR crystal is used in contrast to the ZnSe crystal employed by Basim [125] to simulate the silica wafer sy stem. A nitrogen purged Nicolet Magna 760 spectrometer equipped with liqui d nitrogen chilled MCTA dete ctor was used to carry out FTIR analysis for better accura cy. All the spectra were the result of 512 co-added scans at a resolution of 4 cm 1. The background spectrum for all the experiments was the singlebeam spectrum of the dry silicon crystal. The intensities of the CH2 peaks of the polyurethane were recorded in the wavenumber range of 3000 cm 1. The asymmetric and symmetric stretches of CH2 at wavenumbers of 2927 and 2855 cm-1 respectively were analyzed. The contact area is expected not to depend on pad-wafer relative velocity, hence they are held stationary. The slurries were simulated by using solutions of varying pH with no other chemi cals (passivating agents) or particles added. This is valid especially in silica CMP as no passivating chemicals are generally added in
PAGE 78
66 slurry formulation. The absence of particle s reduces the errors due to scattering. The water FTIR/ATR signal is subtracted from pad spectrum (both at same pH) to get the desired signature spectrum (Fig 3-6). A representative piec e of pad sample (IC-1000) soaked in given solution for one hour was plac ed on the (Si) ATR crystal and loaded at 4, 7 and 10 psi in the wet state under solution. The solutions studied are nanopure water adjusted to pH 4, pH 9 and pH10.5 with either 0.5M HNO3 or 0.5M KOH. Silica CMP is generally performed at slur ry pH 10.5, as the silica solubility increases drastically after pH 10. Thus there should be minimum silica wafer modification or modified layer formation at pH 4 and pH 9. Whereas, pH10.5 should lead to considerable surface dissolution of s ilica. These conditions are chosen, as all CMP process variables except pad properties are suppose d to be constant in pH range of 4 to 9, and increased surface modification (dissolu tion) at pH10.5. The typical absorbance spectra obtained for a given specimen at different applied loads is shown in Fig. 3-7. The signal intensities at symmetric and asymmetric stretches are averaged to calculate the contact area. The results obtaine d for different pH and dry cond itions indicate that contact area increases linearly with applied pressure for all conditions (Fig. 3-8). The normalized plot shows the accuracy associated with the measurement technique, which is consistent for all the conditions used. The linear depende nce of pad-wafer cont act area on applied load validates the use of Greenwood-Williamson theory in the CMP model. Pad-wafer contact area in the wet state was measured to be around 2-3 times th e contact area in the dry state (Fig. 3-9). This implies that the re moval rate calculated w ould be 2-3 times less if the dry contact area is used in the model. There is no discernible change in pad-wafer contact area under different pH conditions due to data scatter, possibly due to uncertainity
PAGE 79
67 associated with subtraction of water spectru m. Utilizing the value of pad-wafer contact area obtained and the Greenwood-Williamson assumption, pad asperity modulus is derived which is used in calculation of pad asperity stress distribution. Fig. 3-6. Analysis of the FTIR-ATR spectra. Fig. 3-7. Absorbance spectra of pad (in pH 4) at different applied pressures.
PAGE 80
68 0.5 1 1.5 2 2.5 3 357911 Pressure (psi)Normalized Contact Area Dry pH 4 pH 9 pH 10.5 Fig. 3-8. Contact area vari ation with applied load in dry and wet conditions. Fig. 3-9. Contact area in dry and we t systems in various pH conditions.
PAGE 81
69 Atomic force microscopy (AFM) Contact area increase in the wet state is attributed to changes in the mechanical properties of the pad. Consequently, surf ace mechanical properties of pad were investigated using a colloidal probe atomic fo rce microscope (AFM), where a particle is attached to AFM tip to measure pad prope rties. A Digital Instruments Dimension 3100 AFM equipped with a liquid cell adapter was used. Particle indentation into pad was used to assess pad surface mechanical properties at pH levels of 4, 9 and 10.5. It was realized that a stiff can tilever is needed to probe the surface mechanical properties of the pad. Hence a tapping mode cantilever was modified by cutting the tip area with focused ion beam (FIB) yielding a cantile ver with a stiffness of 75 N/m as compared to 30 N/m before the modification. A 7m silica particle was attached to the tip of cantil ever for measuring mechanical properties of the pad. Nanoindentation measurements were carried out using the modified AFM tip shown in Fig. 3-10. The pad surface was scanned with the colloid probe to find a relatively flat surface before indentation data was measured. Th is is expected to minimize the effect of surface asperities or roughness on the surface mechanical analysis. Typical indentation data are shown in Fig. 3-11. Only the appro aching curve is used to investigate the pad surface mechanical properties. The results from AFM are summarized in Fig. 3-12. The indentation data presented in Fig. 3-12 indicates that the pad surface mechanical response is markedly different in dry and wet systems. In dry pad case, the force on AFM probe increases with indentat ion depth at a constant pace without any change in stiffness. Whereas, for wet cases at all pH levels employed, a dramatic change in slope is observed. The indentation data is not reproducible over en tire pad surface, as
PAGE 82
70 the pad surface is not perfectly smooth due to asperities. But all the indentation data in wet state showed the stiffness change in contrast to the dry data. When the indentation data is aligned such that all curves match th e dry pad behavior, it can be seen that the pad surface is affected by the solution pH which results in different mechanical response compared to the bulk material (Fig. 3-13). The moduli of pad surf ace and bulk shown in Fig. 3-12 were estimated utilizing Hertz theo ry. The modulus of dry pad is atleast 3-5 times the pad modulus in dry state indicating pad swelling. Attached Particle Attached Particle (a) Silica particle 7 m Silica particle 7 m (b) Fig. 3-10. Colloidal probe AFM tip (a) Mo dified tapping mode cantilever, dotted line shows the region of the cantilever remove d by FIB. (b) Silica particle attached to the modified cantilever.
PAGE 83
71 Fig. 3-11. Force-distance curve indicating softer modified layer on the pad surface. The surface modification of polyuretha ne is not entirely unexpected. The polyurethane polishing pads possess both -C OCand -COOCfuncti onal groups in the main chain structure which are susceptible to hydrolytic attack [50], especially in alkaline media [50]. Disruption of hydrogen bonding is re ported for pad samples immersed in DI water for 5minutes [97]. Schematic of hydr ogen bond disruption is shown in Fig. 3-14. This hydrolytic attack can explain the pa d surface mechanical property behavior. The presence of surface modification on pa d surface or pad swelling could impact the abrasive trapping ability of the pad aspe rities along with the load transmitted to the particles during polishing depending on part icle size. The change in contact area observed in wet state indicates that depending on slurry chem istry, the pad-wafer contact area could be effected, directly influencing the material rate. Hence, when different slurry chemistries are used in a CMP process, the CMP model needs to incorporate the slurry effect not only as a modified layer on wafe r surface, but also as change in pad-wafer
PAGE 84
72 contact area and abrasive tr appability, which has not been accounted for in past CMP modeling efforts. Fig. 3-12. Force-indentation curves from AFM for different conditions. All curves are offset such that they match in the top portions to clearly show the evidence of a softer layer on pad in wet systems. (a) (b) Fig. 3-13. Schematic of pad surface as dedu ced from AFM data. (a) Dry state. (b) Wet state.
PAGE 85
73 (a) (b) Fig. 3-14. Schematic of polishing pads (a ) Structure showing hydrogen bonding between neighboring polymer molecules, (b) Water can form hydrogen bonds with polymer molecule by disrupting hydr ogen bonds between the polyurethane molecules [127]. Pad Asperity Height Distribution A quantitative understanding of the padwafer and pad-particle interface is fundamental to a comprehensive CMP model. The pad asperities are assumed to remain constant through the planarization of each wafer. It is known that the asperities involved in polishing wear out changing in height over time, wh ile the shorter asperities, those that do not come into contact with the wafer, re tain their original height. Thus the pad roughness or the asperity height distribution changes with polishing, if pad conditioning is not performed properly. The surface r oughness of pad is known to be directly proportional to the polishing rate [128-130]. Di fferent types of pads have different asperity height distributions, though they might have similar average surface roughness. In PERC I, pad asperites were assumed to be s pherical and of same si ze with pad asperity height averaged out to obtain the pad asperi ty stress, and pad-wafer contact area. This implies that the pad asperities are not only assu med to be equal in size and shape, but also of equal height. This imposes the pad to have contact spots equal in size, experiencing
PAGE 86
74 same contact pressure. Thus, all the particle s trapped in these cont act spots have same indentation into the wafer surface leading to equivalent material removal under each asperity. Thus the previous PERC model was equivalent to having a single asperity model with no effect of aspe rity height distribution. De pending on the CMP system and desired planarization criteria, different pads are used in practice. Therefore, it was considered necessary to include the asperity height distribution in the revised model (PERC II). Due to constant asperity height only an average indentation depth was derived in the previous model (PERC I) to indicate surface roughness of the polished wafer. When asperity height di stribution is used, asperities at different heights lead to different indentation depth, resulting in information about distribution of scratches (indentation depths).This data can be used to estimate the surface roughness as root mean square roughness (RMS). To incorporate the asperity distribution, Gaussian distribution of asperities is assumed, as the reported literature cites Ga ussian distribution [ 116]. The pressure at contacting asperities were calcu lated individually to estima te the indentation depth of trapped particles. Abrasive Particle Size Distribution The role of slurry in CMP is two-fold. Th e chemicals in the slurry are needed to react with wafer surface resul ting in formation of a surface layer which is amenable to enhanced material removal. The abrasives in the slurry on the other hand, remove this modified surface layer causing material removal. The effect of chemicals is attributed to surface layer of different mechanical propertie s in PERC I model. Whereas, the effect of abrasives is considered by assuming average si ze of the particles. The monosized slurry assumed in the PERC I model is an ideal cas e, since real CMP slurries are seldom
PAGE 87
75 monosized (Fig. 3-15). Oxide po lishing slurries are commonly purchased in concentrated form and diluted on-site to minimize shi pping and labor costs. Some mult-component tungsten and copper polishing slurries must be combined shortly before use because of their short postmix lifetime. The slurry stabil ity can change after it is combined with the oxidizer or due to dilution leading to polydispersity. Fig. 3-15. Particle size distri bution of silica slurry [118] Sometimes mixed abrasive slurries are used [32], which invariably exhibit polydisperse particle size distribution. To acco unt for the particle size distribution effect, PERC I model was modified to incorporat e slurry abrasives with Gaussian size distribution PERC II Model Development Conceptually, the difference between the PERC I and PERC II models can be illustrated as shown in Figs. 3-16 and 3-17. The PERC I model assumes spherical pad asperities of equal height all over the pad (Fig. 3-16a). As the slurry is assumed to be
PAGE 88
76 monosized, each asperity traps equal number of particles in CMP resulting in equal removal rate by each asperity and each particle(Fig. 3-16b). (a) (b) Fig. 3-16. Schematic of pad, slurry and wa fer in PERC I model. (a)Before contact. (b) During CMP. Whereas in the PERC II model (Fig. 3-17a ), the spherical pad asperities are assumed to have a height distribution as obser ved experimentally [116]. As the slurry is assumed to be polydisperse, each asperity trap s different number of particles of different particle sizes in CMP according to pad asper ity height and particle distribution (Fig. 317b). Thus the PERC II model includes rea listic contributions of pad and wafer characteristics compared to the PERC I model. The model formulation is detailed in the following section.
PAGE 89
77 (a) (b) Fig. 3-17. Schematic of pad, slurry and wafer in PERC II model. (a) Before contact. (b)During CMP Real contact area (Areal) between pad and wafer due to applied load (L) is given by: dz z h z NA Ah o real (3.1) dz z h z E NA Lh o 2 3 2 1 *3 4 (3.2) 2 22 z exp 2 1 z (3.3) where N is the number density of the pad asperities (i.e., number of asperities per unit area), Ao is the total area of th e flat wafer surface, and h is the distance between wafer
PAGE 90
78 surface and the pad asperity averag e plane under the given conditions. E* is the composite modulus between the two surfaces in contact, defined as w w p pE E E2 2 *1 1 1 (3.4) where, E and are the Youngs modulus and the Po issons ratio, and the subscripts p and w represent the pad and the wafer, respectiv ely. Equations (3.1) and (3.2) are derived from Hertz theory as in Greenwood-Williams on model [77, 117], yielding contact area and load at a given asperity of height z to be: Areal(z)= (z-h) (3.5) L(z)=(4/3)E*1/2(z-h)3/2 (3.6) where h is calculated from Eq. 3.2, since applied load and pad properties are known. The underlying assumption for Eq. 3.2 is that the applied normal force on the wafer is supported entirely by the pad asperities. As was mentioned previously, experimental measurements indicate that the hydrodynamic pressure is negligible under typical CMP operational conditions [69]. From equations 3.5 and 3.6, contact pressu re and contact area at a given asperity height can be calculated assuming that the pa d asperity distributi on is known. The pad asperity distribution can be determined from routin e surface roughness measurement techniques like optical and force profilomet ers or atomic force microscopy (AFM). Contact pressure according to Hertz theory, on an asperity of height z is given by: 2 1 2 1 *3 ) ( 4 ) (h z E z P (3.7)
PAGE 91
79 Plastic Deformation of Wafer Surface When an abrasive particle is indented into the wafer surface, elastic and plastic deformation of wafer and abrasive particle occur. However, in CMP, slurries are designed such that the chemistry changes th e modulus of the wafer surface for enhanced material removal. In the current model, partic les are assumed to be rigid such that they do not deform during polishing. When the indent ation of abrasive particle is beyond the modified surface layer, the substrate indenta tion could yield either elastic or plastic response depending on the substrate, for exampl e in case of silica, the deformation will still remain plastic whereas for copper, it would be elastic deformation followed by plastic deformation. Therefore, in material removal calculation, indentation into ductile wafers is not considered to cause material re moval, as it leads to change in shape only, whereas in the case of brittle material like si lica, cracks are developed leading to material removal. Slurry Characteristics Particle Size Distribution The abrasive particles in the slurry are assumed to be spherical in shape with a given size distribution (r).As mentioned earlier, particle size distribution (r) was assumed to be Gaussian, given by:2 22 exp 2 1 ) (r r rr r (3.8) where r is the mean particle radius and r the standard deviation. Material Removal by Single Particle The overall material removal rate (RR) is assumed to be the multiple of the removal rate by a single part icle and the number of par ticles trapped between pad asperities and the wafer:
PAGE 92
80 RR = (1/A0) n(r,z) RRv(r,z)drdz (3.9) where n(r,z) is the number of slurry particles an chored between pad asperity and the wafer and RRv(r,z) is the volumetric removal rate by a single particle of radius r trapped by asperity of height z. The overall removal rate, RR is typically described in terms of the thickness of the wafer surface layer removed per unit time; hence the division by the nominal wafer area Ao. Schematic illustration of single particle abra sion is given in Fig. 3-18. The material removal is viewed as a sliding-indentation process in which the volumetric removal rate by a single abrasive particle is given by RRv(r,z) = Ai(r,z)V (3.10) Where Ai(r,z) is the cross-sectional area of the particle indented into the wafer under pressure (hashed area in Fig. 3-18), and V is the relative velocity between the pad asperity and the wafer. Fig. 3-18 Schematic of single particle ab rasion showing the removal of chemically modified surface layer, (a). t ; (b). > t with soft wafer materials.[72]
PAGE 93
81 Assuming that the hard particle maintains its spherical shape, r z r a r z r r z r a z r Ai) ( arcsin ) ( ) ( ) (2 (3.11) where r,z is the indentation depth and a(r,z) is the radius of the circular indentation on the wafer surface caused by the particle of radius r trapped under asperity of height z. Because r,z is very small compared to r, a (r,z)/r is also very small and Equation 3.11 can be linearized about a(r,z) /r as already discussed: r z r O z r z r a z r Ai) ( 1 ) ( ) ( 3 4 ) ( (3.12) Furthermore, since a2(r,z)=(2r (r,z) 2(r,z)) and r,z) is much smaller than r, Equation 3.12 can be rewritten as r z r a Ai3 ) ( 2 z) (r,3 (3.13) Once the indentation depth, r,z) (or a (r,z)) is known, the vol umetric removal rate by a single particle can be calculated fr om Eq. 3.13. It should be noted that and a are dependant on particle radius a nd height of the trapping asper ity. Removal rate per particle is given by: r z r a z r RRv3 ) ( 2 ) (3 V (3.14) Depending on the relative magnitude of the indentation depth ( (r,z)) as compared to the surface layer thickness (t), the material removal may be classified into two different regimes as depi cted in Fig. 3-18: (a) (r,z) t and (b) (r,z) > t and can be calculated using force balance as follows.
PAGE 94
82 Considering that the pad is sufficiently hard as compared to the modified layer, the value of b in Fig. 3-18 can be clos e to the particle radius R, 2 / 1) ( ) ( eB z P r z r a when (r,z) t (3.15) and 2 / 1 2) ( 2 ) ( ) ( w e wB rt B B z P r z r a when (r,z) > t (3.16) where, Be is the effective Brinell hardness of the chemically modified surface layer. In the following section, the number of part icles in contact with the wafer surface is estimated to determine the overall removal rate. Number of Active Abrasive Particles When the wafer is pressed against the pad and is in sliding motion relative to the pad, the slurry particles can be entrapped only in the region where the local distance between the wafer and the pad asperity is smaller than the particle diameter, 2 R For a slurry consisting of particles with Gaussi an distribution, mean particle size (Rm) is used to calculate the number of particles that can be trapped on the pad aspe rity of height z: N(z) = 2RmAreal(z) n (3.17) where Rm is the mean abrasive particle radius and Areal(z) is the real contact area between the wafer surface and the pad asperity as given by eq. 3.5. n is the number density of the particles in the slurry (i.e., number of pa rticles per unit slurry volume), that can be calculated from the solid loading Ca (wt%) of the slurry particles as dr r r Ca s a n) ( 4 33 (3.18)
PAGE 95
83 where a and s are the densities of the particle a nd the slurry soluti on, respectively and (r) the slurry particle size distribution. Substitution of Eq. 3.18 into 3.17 gives dr r r z A C R z Na real s a m) ( ) ( 2 3 ) (3 (3.19) From eq. (3.19), the number of trapped particles of radius R is given by, n(R,z)=N(z)* (R)= dr r r r z A C Ra real s a m) ( ) ( ) ( 2 33 (3.20) Estimation of the Overall Removal Rate The overall material removal rate can be determined by following these steps: h is calculated using eq. 3.2. Contact pressure P(z) at a given asperity is calcul ated along with the real contact area Areal(z) of that asperity with wafer. Total number of particles N(z), that are trapped are determined, from which number of particles of given radii, n(r,z) are derived. From P(z), removal rate by single abrasive of radius r is calculated which is multiplied with n(r,z) to give removal ra te by that particle size. This process is repeated to account for all the particles of diffe rent radii trapped on that asperity yielding material removal rate by particles trapped on a given asperity. It should be noted that there are (z)NA0 asperities at a given height which will each contribute the same material removal.
PAGE 96
84 The above step is repeated for all the pad asperities of va rious heights that are in contact with wafer (determined by h) to give overall removal rate by the following equation: RR = (1/A0) n(r,z) RRv(r,z)drdz RR= hr real r a s a mdz dr r z r a r z z A dr r r N C VR ) ( ) ( ) ( ) ( ) (3 3 (3.21) where, a(r,z) varies as given in eqs.3. 15 and 3.16 depending on particle radius and asperity height. The indentation depth, (r,z) is used to calculate the surface roughness RMS of the polished wafer according to the following equation: n RMSn i i1 2 (3.22) where i is the indentation depth. The overall removal rate predicted by Eq. 3.21 and surface roughness given by Eq. 3.22 depend on various parameters that represent the geometry of the pad and the particle (i.e., and R ), material properties ( E*, Bw and Be), thickness ( t ) of the modified surface layer on wafer, and the relative velocity of the pad (thus the particle) to the wafer ( V ). The PERC II model will be validated with re spect to critical CMP parameters in the next chapter.
PAGE 97
85 CHAPTER 4 MODEL ASSESSMENT The parameters governing the material removal calculation consist of CMP operating parameters, slurry parameters, pad and wafer parameters, as listed in Table 4-1. However, some of these parameters, especia lly the sub-nanometer thick modified layer hardness and thickness can not be determin ed experimentally due to equipment limitations. The remaining parameters are rela tively easy to investigate experimentally for many CMP systems. Hence, modified la yer hardness and thickne ss are used as the fitting parameters in the model. Table 4-1. Parameters incl uded in the revised PERC model. Model Validation In this section, PERC II model predictions are compared with experimentally measured values to assess the reliability of the proposed CMP model. The experimental literature data is also compared with PERC I predictions to assess the advantage of PERC II over PERC I. Though PERC II is a refined model compared to PERC I, the model is still a simplified version of real CMP, mainly in estimating the number of trapped Operating Parameters Pressure Velocity Slurry parameters Particle size distribution Solids loading Abrasive density Solvent density Pad properties Pad modulus Asperity height distribution Asperity radius Asperity density Wafer properties Modified layer thickness Modified layer hardness Wafer hardness
PAGE 98
86 particles and indentation dept h considering an abrupt tran sition between modified layer and wafer substrate. The removal rate values for both models are normalized by a constant to have reasonable modified layer thickness values. Surface roughness data is as produced by the models without any normalizi ng factor. The size distribution of slurries given in the respective literature are fit to the closest Gaussian distribution, which is used in the model. Silica CMP: Micron Sized Slurries The polishing results of silica by silica abra sives as reported by Choi et al. [131] are considered here. Micron sized CMP slurri es were formulated by dispersing sol-gel silica particles of different sizes from Ge ltech Corporation in deionized water and adjusting pH values to 10.5. The average diamet er of the slurry particles were 0.2, 0.5, 1.0 and 1.5 m. The size distribution of the slurries is plotted in Fig. 4-1. A 2 m thick silica layer was deposited on p-type silic on wafer by plasma-enhanced chemical vapor deposition. 1.5 x 1.5 in silica samples were used to mimic the wafers. A Struers RotoPol31 tabletop polisher was used with IC 1000/Suba IV stacked pad. The following conditions were applied: down pressure of 3. 5 psi, slurry flow rate of 100 ml/min, and rotation speed of 150 rpm (110 cm/s). The pol ishing rates were determined by measuring oxide film thickness using variable angle sp ectrometry ellipsometry and surface quality was assessed by atomic force microscopy (AFM). The polishing rate was investigated as a function of particle size and solids loading. The removal rate decreased with increasing pa rticle size (Fig. 4-2). The parameters used are: wafer Hardness=8 GPa, modified la yer Hardness = 450 MPa, modified layer thickness = 7.2 and Normalization factor = 100. The PERC II model correlates well
PAGE 99
87 with the experimental values than the PERC I model. The reported literature belongs to the substrate indentation regime. The model pr ojects that enhanced removal rate can be obtained by using smaller sized slurries The surface roughness on the other hand, increases with particle size on an average for all the slurries consid ered (Fig. 4-2b). This is not surprising as the larger size particles have deeper indentati on depths increasing surface roughness. Both the models predict the average trend in surface roughness with increasing particle size. Fig. 4-1. Slurries used in oxide polishing [131]. Silica CMP: Nanosilica slurries The slurry sizes in the earlie r study spanned from 0.2 to 1.5 m, which is well above the normal slurry size range (<300 nm). Considering the nanosize slurries of 50150nm mean particle sizes, CMP behavior of silica polis hing by silica abrasives is investigated further [116]. Slurries with mean particle diameters of 50, 80, 110 and 140nm were used to polish 100mm diameter, p-type polished silicon wafers that contained a thickness of 1.3 m thermally grown oxide. The silica particles were
PAGE 100
88 precipitated from a collo idal solution. The polishing was done at three different levels of pressure and relative pad-wafer velocities : 18kPa, 0.4m/s; 32kPa 0.6m/s; and 45kPa, 0.9m/s. The parameters used in the model are: wafer Hardness=8 GPa, modified layer Hardness = 250 MPa, and modified layer thic kness = 9.3 The removal rates for the given set of pressure and velocity with differe nt particle sizes is s hown in Fig. 4-3. The material removal rates with slurry mean sizes of 10 and 20nm were ignored, as the authors [116] noticed considerable particle staining on the polishe d surfaces indicating agglomeration. The PERC II model corresponds with the experimental results better than the previous PERC I model. Each set of mode led removal rates for given pressure and velocity were normalized individually w ith normalizing factor ranging from 185-200. The discrepancy in the removal rate at particle size 50nm between experimentally observed values and modeled values can be attributed to agglomeration, which is suggested by the authors for that size [116]. (a) Fig. 4-2. Silica polishing by sl urries in the micron range. (a) Removal rate. (b) Surface quality
PAGE 101
89 (b) Fig. 4-2. Continued. Fig. 4-3. Silica polishing by colloidal nanosilica. The surface roughness of wafers polished wa s reported to be 6 on an average. The PERC II model predicts the surface roughness to be from 5.8 6.2 which is in good agreement with the experimental result s. The PERC I model predicts an average indentation depth of 5.5 to 6 wh ich is also in good correlation.
PAGE 102
90 As the model correlated well with th e micron and submicron sized slurries, nanosized slurries in the low-end regime ( 10-100nm) are considered further. Colloidal particles of size 12, 25,45, 75nm were used to polish thermal oxide (silica) wafers with potassium iodate (KIO3) in the slurry [118]. The remova l rate was reported to increase with particle size, r each an optimum and then decrease wi th particle size (Fig. 4-4). The parameters used are: wafer Hardness=8 GPa, modified layer Hardness = 200 MPa, modified layer thickness = 5.1 and No rmalizing factor=330. The PERC I model predicts the removal rate to remain consta nt with particle size, whereas the PERC II model predicts the optimal particle size at which enhanced material removal can take place. The surface roughness predicted is higher than the reported values. (a) Fig. 4-4. Silica polishing by nanosized slur ries. (a) Removal rate. (b) Surface quality.
PAGE 103
91 (b) Fig. 4-4. Continued. Copper CMP As the model is found to correlate well w ith the brittle wafer polishing, metal CMP process is now considered to review the comprehensive nature of the model. Copper wafers were polished with particles of size 100, 200 and 300 nm [132]. The chemicals used were 5 wt% H2O2 and 1 wt% glycine with pH adjusted to pH 4. The parameters used are: wafer Hardne ss=1.75 GPa, modified layer Hardness = 300 MPa, modified layer thickness = 15 a nd Normalizing factor=7. The PERC I model predicts removal rate independe nt of particle size, while th e PERC II correlates well with the observed values (Fig. 4-5). The opposing trend of surface roughness between experimental and observed values is surpri sing. The discrepancy in surface quality predictions could indicate pr esence of few bigger sized part icles or agglomeration which can result in high surf ace roughness than predicted [125] The presence of larger size particles or slurry agglomeration, is however not mentioned by the authors.
PAGE 104
92 Tungsten CMP Bielman et al., [133] have conducted CMP of tungsten with alumina slurries of size distributions shown in Fig. 4-6. The slurry contained K3Fe(CN)6 as the chemical additive with pH adjusted to pH 3. The rem oval rate at 15wt% solids loading is shown in Fig. 4-7. The parameters used are: wafer Hardness=6.5 GPa, modified layer Hardness = 1.5 GPa, modified layer thickness = 9 a nd normalization factor = 23. The PERC II model correlates well with the experimental removal rate values compared to the PERC I model (Fig. 4-7a). The surface roughness valu es predicted by PERC I and II are both higher compared to the observed values, but the surface roughness trend with particle size matches very well. 0 100 200 300 400 500 50100150200250300350400 Mean Particle size (nm)Removal Rate (nm/min) Experimental PERC II Model PERC I Model (a) Fig. 4-5 Copper CMP as a function of particle size. (a) Removal rate, (b) Surface quality.
PAGE 105
93 (b) Fig. 4-5 Continued. Fig. 4-6. Alumina slurry size distributions [133].
PAGE 106
94 0 200 400 600 800 1000 05001000150020002500 Mean Particle Size (nm)Removal Rate (nm/mi Experimental PERC II Model PERC I Model (a) 0 5 10 15 20 25 05001000150020002500 Mean Particle Size (nm)Surface Quality (Angstroms) Experimental PERC II Model PERC I Model (b) Fig. 4-7. Effect of particle size on tungst en CMP. (a) Removal rate, (b) Surface quality. Model Predictions The better match of PERC II model predictions than the PERC I model indicates that the refinements made to the PERC I model assumptions are indeed significant and add to the predictive capability of the model. In the following sections, predictive capability of PERC II model is explored, comparing with the PERC I model. The
PAGE 107
95 enhanced performance of the proposed m odel (PERC II) is compared with PERC I predictions considering the case of a brittle wafer as in Silica CMP. The parameters used for comparison are listed in Table 4-2. Table 4-2. Parameter values used for comparing PERC I and II models. Pad modulus 60 MPa Pad asperity density 2 x 108/m2 Mean pad asperity radius 50 m Pad asperity height distribution 6.7 m Pressure 6.5 psi Velocity 1 m/s Particle size 200 nm Particle size distribution r 16 nm Abrasive density 2.2 gm/cc Solids loading 10 wt% Wafer hardness 8 GPa Layer hardness 600 MPa Layer thickness 10 Removal Rate as a Function of Partic le Size for a Given Solids Loading: The comparison of the two models with varyin g particle size is shown in Fig. 4-8. As discussed earlier (in chapter 3), materi al removal rate as per PERC I model is independent of particle size. The PERC II model, however, predicts an increase in removal rate with particle size during modified layer re moval, and as the particle indentation proceeds into wafer substrate, the removal rate decr eases smoothly with particle size, as also observed experimentally. Decrease in re moval rate with particle size in wafer indentation is caused by the substr ate being harder compared to the modified layer. The PERC II model differs with the PERC I model mainly in estimation of the modified layer removal and transition regi mes and not considerably in the substrate removal regime. The smooth transition from mo dified layer removal regime to substrate removal regime is attributed by inclusion of pad asperity distribution into the model. Whereas, the removal rate increase with par ticle size in the modified layer removal
PAGE 108
96 regime is due to the inclusion of slurry pa rticle size distribution and strongly depends on it. Fig. 4-8. Comparison between PERC I and PERC II models with constant solids loading of different particle sizes. (a) Removal rate. (b) Surface roughness. The other feature of the PERC II model is the estimation of surface quality as root mean square surface roughness and maximum scratch depth. The PERC I model on the other hand, gave an estimation of surface roughness through average indentation depth. 0 100 200 300 400 500 0100200300400500 Mean Particle Size (nm)Removal Rate (nm/min ) PERC II Model PERC I Model (a) 0 2 4 6 8 10 12 0100200300400500 Mean Particle Size (nm)Surface Roughness RMS (Angstroms) PERC II Model PERC I Model ( b )
PAGE 109
97 The PERC II model, however gives the scra tch (indentation) depth under each contacting pad-asperity due to each trapped particle size. This information is used in calculation of the surface roughness (RMS) as de scribed in the previous chapter. The PERC I model yielded an average scratch depth whose de pendence on particle size suddenly changed when the particle was large enough to indent pa st the modified layer into the substrate. PERC II also reveals that particle indentation into the substrate takes place at a smaller mean particle size as compared to that predicted by PERC I model. This is expected, as some of the larger partic les in the slurry indent into the substrate, while most others indent into the modified layer alone keeping the average particle size lower. Indentation in to the wafer substrate below the modified layer is not a desirable feature, as it negates the purpose of CMP. But, as long as the substrate removal is minimal with low indentation depths, it can be tolerated. Effect of Modified Layer Thickness The chemical reaction between slurry chemicals and wafer surface leads to the modified layer, whose thickness and mechani cal properties are strongly dependent on the type of chemicals present in the slurry. The thickness of the modified layer in an optimized CMP process should be such that the particle indents completely into the modified layer and not into th e substrate below. However, in a slurry with particles of various sizes, not all particle sizes can indent/remove the entire modified layer. From Figs. 4-9 a and b, the transition from modifi ed layer to substrate indentation regime occurs around 8 which corresponds to 200nm pa rticle size. Therefore, as the modified layer thickness decreases below 8 keeping a ll the other parameters constant, the slurry particles indent into the substrate and the removal rate decreases. Whereas, when the modified layer thickness increases beyond 8 the indentation becomes more and more
PAGE 110
98 limited to modified layer alone. Thus, after a critical thickness, the removal rate remains constant, even if the modified layer thickness increases. This indicates that the chemical aggressiveness of the slurry should not focu s on forming a thicker layer, but on forming modified layer faster of desired thickness. 0 100 200 300 400 500 05101520 Modified Layer Thickness (Angstroms)Removal Rate (nm/min ) PERC II Model PERC I Model (a) 0 1 2 3 4 5 6 7 05101520 Modified Layer Thickness (Angstroms)Surface Roughness RMS (Angstroms) PERC II Model PERC I Model (b) Fig. 4-9. Model comparison with varying modi fied layer thickness. (a) Removal rate, (b) Surface quality.
PAGE 111
99 Effect of Modified Layer Hardness As the modified layer hardness increases, less and less indentation of the particles take place, resulting in lower material removal rates (Fig.s 4-10 a and b). Thus, if there is no surface modification due to slurry chemicals in silica, minimal removal rate will be observed. As 200nm particle was in the transi tion region, with incr easing modified layer hardness, most of the particles indent onl y the modified layer. The transition from modified layer to substrate indentation is depicted more clea rly by the PERC II model than by the abrupt transition predicted by the PERC I model. Combining the model predictions and valid ation of metallic and dielectric CMP systems, the PERC II model can be used as a CMP parameter selection guide for emerging CMP systems and to optimize the current CMP processes. The fitting parameters, modified layer hardness and thicknes s are in the range of expected values for the respective CMP systems. The thickness of the modified layer determines the optimal particle size predicted, whereas the modifi ed layer hardness in comparison to wafer hardness determines the sharp decrease in re moval rate after transition from modified layer to substrate removal. The increase in re moval rate with particle size in the modified layer removal regime is influenced solely by the abrasive particle size distribution. The normalizing factor is found to be sp ecific to the CMP system and strongly dependent on the toolset used. The normalizing values of 185-200 for three sets of pressure and velocities used in silica CMP by nanosilica slurries [116], indicate that the normalizing value is influenced strongly by th e toolset used. This is expected, as the ability of pad to trap the pa rticles is strongly dependent on the toolset configuration. The normalizing factor is also e xpected to depend on type of wafer being polished and the slurry used. For example, if pa rticles are attracted to wafer due to electrostatic attraction,
PAGE 112
100 the number of trapped particles will be mu ch less compared to the system, where particles are repelled by the wafer surface charge. The next chapter summarizes the findings in this study and future work is s uggested based on the insights obtained through the study. 0 250 500 750 1000 00.250.50.751 Modified Layer Hardness (GPa)Removal Rate (nm/min ) PERC II Model PERC I Model (a) 0 2 4 6 8 10 12 00.250.50.751 Modified Layer Hardness (GPa)Surface Roughness RMS (Angstroms) PERC II Model PERC I Model (b) Fig. 4-10. CMP performance comparison with varying modified layer hardness. (a) Removal rate, (b) Surface quality.
PAGE 113
101 CHAPTER 5 SUMMARY AND SUGGESTIONS FOR FUTURE WORK The device shrinkage in microelectro nics industry nece ssitates chemical mechanical polishing and continues to de mand stringent performance of CMP. The material removal mechanism needs to be unders tood with the relative contributions from particle-wafer, pad-wafer, a nd pad-particle interfaces and the reaction of slurry with wafer and pad materials. In this study, a mode l is developed to predict process parameters for designing an optimal CMP process. Most of the CMP modeling efforts concentr ated mainly on explaining the Prestons equation. Therefore, most of them were focu sed on product of pressure and velocity (PV) as the operating parameter and not on other important parameters like slurry and pad variables. The PERC II model is devel oped based on PERC I model [72]. The assumptions in PERC I model were refined to include particle size distribution of slurry and pad asperity height dist ribution. Though recently a model developed by Luo [27] incorporated particle size di stribution, part of the distri bution was neglected and hence doesnt incorporate the proce ss variables appropriately. The pad-wafer contact area dependence on applied load is generally assumed in many CMP models with no basis of a systematic study. The material removal rate is directly proportional to the pad-wafer cont act area. Therefore, the magnitude of pad contact area is significant in modeling the process. The earlier studi es focused on contact area in dry state and the results reported were not consistent. The contact area was investigated in dry and wet states in the current study. Contact area in dry state was
PAGE 114
102 studied using optical microscopy. The change in contact area due to presence of aqueous environment and different pH conditions is i nvestigated using Fourie r transform infrared spectroscopy. The contact area was found to be 2-3 times the contact area in dry state. The pad surface mechanical properties were al so investigated at different pH conditions using Atomic force microscopy. The slurry was found to react with pad surface layer (~50nm thick), modifying its mechanical prope rties. Both the padwafer contact area and contact stresses are strongly dependent on pad asperity height distri bution. To incorporate the effects of contact area and contact stresse s, Gaussian asperity height distribution of asperities is incorporated into the PERC II model. The slurry used in commercial CMP is se ldom monosized. The size distribution of the particles can strongly aff ect the process. Hence, the slurry abrasive sizes were assumed to be Gaussian in distribution were incorporated into the PERC II model. The developed model correlates well with expe rimental data for oxide and metal CMP systems. The model can be used as a sele ction tool for choosin g the proper parameter values optimizing the CMP process. For optimizat ion of the process, a sensitivity analysis can be performed to narrow down the parame ters that are of utmost importance. An example is shown here of silica CMP system with the parameter ranges shown in table 51. SOBOL analysis was performed with 4608 data sets (using SimLab software) assuming particle size distributi on to be Gaussian with stan dard deviation equal to the one-fourth the mean particle size. The result s revealing the first order and interaction effects of parameters on removal rate are s hown in Fig. 5-1. The sensitivity analysis indicates that the particle size has the signifi cant effect on removal ra te compared to all the other parameters. The particle size dete rmines if the material removal will be in
PAGE 115
103 modified layer regime or will extend to s ubstrate indentation regime. Depending on the indentation regime, the hardness of the ma terial being removed strongly effects the material removal. Table 5-1. Process parameter ranges used in sensitivity analysis. Factor Limits Pressure 2-8 psi Mean Particle Size 10-500 nm Asperity Radius 20-100 m Asperity Height S.D 1-20 m Asperity Density 0.0220 x 109/m2 Pad Modulus 20-200 MPa Modified Layer Hardness 200-500 MPa Modified Layer Thickness 5-10 Fig. 5-1. Sensitivity analysis of silica CMP system. The reported findings in this study pr ovide insight into the fundamental understanding of the polishing mechanism and can give guidelines for optimizing any given CMP system.
PAGE 116
104 Suggestions for Future Work The insight provided by the current study in to the CMP process in general can be used to conduct further studies to bett er understand the process from various perspectives. The study focuses on slurry with ri gid abrasive particles. With the advent of low-k dielectric polishing, so fter abrasives are being us ed. Inclusion of non-rigid abrasives into the model will facilitate better exploitation of the process potential and faster adaptation of emerging CMP systems to the industry requirements. Recently, many commercial slurries are de signed with surfactants or particle surface modification for slurry stability. Add ition of these extra factors can significantly effect the CMP process. For example, it is shown by Basim [125] that polishing performance is degraded with addition of certain surfactant cocentration to the slurry. These steric effects or electro static effects due to ionic sp ecies in the slurry should be included in the model for comprehensiv e understanding of the effects involved. Similarly, effects of pH and ionic speci es on CMP processes [134-137] can also be included. The current model assumes that there are no stresses involved in the wafer, which could affect the CMP process. But, it has b een shown by McGarth et al. [138] that if the stresses induced in the wafer due to mate rial deposition and ot her processes are not relieved, the CMP process could be affected Inclusion of such a stressed wafer could help in understanding the e ffect of deposition process on CMP process performance. One of the experimental challenges faced in this study was the measurement of the modified layer properties. Nanoindentation and AFM were used to investigate the modified layer on the silica and copper surfaces. But, the instruments were not sensitive
PAGE 117
105 enough to detect the modified layer propertie s quantitatively. It is suggested that a diamond tipped AFM cantilever with a narrow ne edle type tip be used to sense the modified layer properties, wit hout any interference from the substrate. This data can be used to incorporate gradual transition between modified layer and substrate compared to the abrupt transition employed now. The current model assumes formation of a modified layer on the wafer surface which is critical in material removal pro cess. Based on the AFM data of pad surface, even a modified layer on the pad can be a ssumed to assess the trapping capability of abrasives and load transmittal of pad onto the wafer through the abrasives. Developing a model including such a layer on pad will re sult in better predictive capability when different slurry chemicals are used that effect pad properties. The model developed shows predictive capability in the particle scale. Therefore, it can be extended to die and wafer scales to predict the topogr aphy evolution during polishing. Such a model extension would also involve including the se lectivity aspect of the CMP slurry, dishing and erosion.
PAGE 118
106 LIST OF REFERENCES [1] Chang, C.Y., Sze, S. M., ULSI technology. New York: McGraw-Hill 1996. [2] News, E., "IBM, AMD to jointly de velop 65nm, 45nm chip processes." 2003, http://www.reed-electronics.com/e lectronicnews/article/CA269292.html August2005. [3] Ryan, J.G., Geffken, R. M., Poulin, J. R., Paraszczak, J.R., "The evolution of interconnection technology at IBM." IBM Journal of Research and Development 39(4): p. 371, 1995. [4] Borst, C.L., D.G. Thakurta, W.N. Gill, and R.J. Gutmann, "Chemical-mechanical planarization of low-k polymers for advanced IC structures." Journal of Electronic Packaging 124(4): p. 362-366, 2002. [5] Bates, A.K., Rothschild, M., Blomstein, T. M., Fedynyshyn, T. H., Kunz, R. R., Liberman, V., Switkes, M., "Review of technology for 157nm lithography." IBM Journal of Research and Development 45: p. 605, 2001. [6] Fay, B., "Advanced optical lithograp hy development, from UV to EUV." Microelectronic Engineering 61-62: p. 11-24, 2002. [7] Michel, B., A. Bernard, A. Bietsch, E. Delamarche, M. Geissler, D. Juncker, H. Kind, J.P. Renault, H. Rothuizen, H. Schmid, P. Schmidt-Winkel, R. Stutz, and H. Wolf, "Printing meets lithography: soft a pproaches to high-re solution printing." IBM Journal of Research and Development 45(5): p. 697-719, 2001. [8] Wolf, S., Tauber, R., Silicon processing for VLSI era. New York: McGraw Hill 1991. [9] Murarka, S.P., Metallization: Theory and practice for VLSI and ULSI. Stoneham, MA: Butterworth-Heinemann 1993. [10] Zantye, P.B., A. Kumar, and A.K. Sikde r, "Chemical mechanical planarization for microelectronics applications." Materials Science & E ngineering, R: Reports R45(3-6): p. 89-220, 2004. [11] Lee, J.A., Moinpour, M., Liou, H.-C., Abell, T. "Integration challenges for chemical mechanical polishing of cu/low-k interconnects." in Materials Research Society Proceedings San Francisco 2003. [12] ITRS, 2001, http://pub lic.itrs.net, August 2005.
PAGE 119
107 [13] Wolters, P., "Multi-l evel metallization." 2000, http://www.peterwolters.com/cmp/cmpmultilevel.htm August2005, [14] Shon-Roy, L., "Integrated circu it engineering cor poration handbook." 1998, http://smithsonianchips.si.e du/ice/cd/STATUS98/TOFC.PDF August 2005. [15] komatsu, "History of the company." 2003, http://www.komsil.co.jp/en/office/hist.htm August2005, [16] Sematech, "Sematech history." 2004, http://www.sematech.org/corporate/history.htm August2005, [17] Oliver, M.R., "Cmp technology." Springer Series in Materials Science 69(Chemical-Mechanical Planarization of Semiconductor Materials): p. 7-40, 2004. [18] Cooperman, S.S., A.I. Nasr, and G.J. Grula, "Optimization of a shallow trench isolation process for im proved planarization." Journal of the Electrochemical Society 142(9): p. 3180-5, 1995. [19] Steigerwald, J.M., Shyam, P., Gu tmann, M., Gutmann, R. J., Chemical mechanical planarization of microelectronic materials. New York: Wiley 1997. [20] Singh, R.K. and R. Bajaj, "Advances in chemical-mechanical planarization." MRS Bulletin 27(10): p. 743-747, 2002. [21] Patrick, W.J., W.L. Guthrie, C.L. St andley, and P.M. Schiable, "Application of chemical mechanical polishing to the fabrication of VLSI circuit interconnections." Journal of the Electrochemical Society 138(6): p. 1778-84, 1991. [22] Guthrie, W.L., W.J. Patrick, E. Levi ne, H.C. Jones, E.A. Mehter, T.F. Houghton, G.T. Chiu, and M.A. Mury, "A four-level vlsi bipolar metalization design with chemical-mechanical planarization." IBM Journal of Research and Development 36(5): p. 845-57, 1992. [23] Tsai, T.-H., Y.-F. Wu, and S.-C. Yen, "A study of copper chemical mechanical polishing in urea-hydrogen peroxide sl urry by electrochemical impedance spectroscopy." Applied Surface Science 214(1-4): p. 120-135, 2003. [24] Cook, L.M., "Chemical pr ocesses in glass polishing." Journal of Non-Crystalline Solids 120(1-3): p. 152-71, 1990. [25] Vijayakumar, A., T. Du, K.B. Sundaram, and V. Desai, "Polishing mechanism of tantalum films by sio2 particles." Microelectronic Engineering 70(1): p. 93-101, 2003.
PAGE 120
108 [26] Kourouklis, C., T. Kohlmeier, and H.H. Gatzen, "The application of chemicalmechanical polishing for planarizing a su -8/permalloy combination used in mems devices." Sensors and Actuators, A: Physical A106(1-3): p. 263-266, 2003. [27] Luo, J.F. and D.A. Dornfeld, "M aterial removal mechanism in chemical mechanical polishing: Theory and modeling." IEEE Transactions on Semiconductor Manufacturing 14(2): p. 112-133, 2001. [28] Kondo, S., N. Sakuma, Y. Homma, Y. Goto, N. Ohas hi, H. Yamaguchi, and N. Owada, "Abrasive-free polishing for copper damascene interconnection." Journal of the Electrochemical Society 147(10): p. 3907-3913, 2000. [29] Hanazono, M., J. Amanokura, and Y. Kamigata, "Development and application of an abrasive-free polishing solution for copper." MRS Bulletin 27(10): p. 772-775, 2002. [30] Kim, S.-Y., K.-J. Lee, and Y.-J. Seo, "In-situ end point dete ction of the sti-cmp process using a high se lectivity slurry." Microelectronic Engineering 66(1-4): p. 463-471, 2003. [31] Jindal, A., S. Hegde, and S.V. Babu, "C hemical mechanical polishing of dielectric films using mixed abrasive slurries." Journal of the Elect rochemical Society 150(5): p. G314-G318, 2003. [32] Jindal, A., S. Hegde, and S.V. Ba bu, "Chemical mechanical polishing using mixed abrasive slurries." Electrochemical and Solid-State Letters 5(7): p. G48G50, 2002. [33] Anon, "End point detection of mate rial removal in polishing processes." Research Disclosure 322: p. 102, 1991. [34] Fukuroda, A., K. Nakamura, and Y. Ar imoto, "In situ cmp monitoring technique for multi-layer interconnection." Technical Digest In ternational Electron Devices Meeting : p. 469-472, 1995. [35] Tucker, T., "Equipment used in cmp processes." Springer Series in Materials Science 69(Chemical-Mechanical Planarization of Semiconductor Materials): p. 133-165, 2004. [36] Sainio, C.A., D.J. Duquette, J. Steigerw ald, and S.P. Murarka, "Electrochemical effects in the chemical-mechanical polishi ng of copper for integrated circuits." Journal of Electronic Materials 25(10): p. 1593-1598, 1996. [37] Kaufman, F.B., D.B. Thompson, R.E. Broadie, M.A. Jaso, W.L. Guthrie, D.J. Pearson, and M.B. Small, "Chemical-mechanical polishing for fabricating patterned tungsten metal features as chip interconnects." Journal of the Electrochemical Society 138(11): p. 3460-5, 1991.
PAGE 121
109 [38] Li, S.H., R.O. Miller, and Editors, "Chemical mechanical polishing in silicon processing. [in: Semicond. Semimetals, 2000; 63]." 307 pp., 2000. [39] Tsai, T.-H. and S.-C. Yen, "Localiz ed corrosion effects and modifications of acidic and alkaline slurries on coppe r chemical mechanical polishing." Applied Surface Science 210(3-4): p. 190-205, 2003. [40] Iler, J., The chemistry of silica. New York: Wiley 1979. [41] Basim, G.B. and B.M. Moudgil, "Eff ect of soft agglomerates on cmp slurry performance." Journal of Colloid and Interface Science 256(1): p. 137-142, 2002. [42] Singh, P.K., "Dispersion of nanopart iculate suspensions using self-assembled surfactant aggregates.", in Materials Science and Engineering 2002, University of Florida: Gainesville, FL. [43] Kim, J., U. Paik, Y.-G. Jung, T. Kat oh, and J.-G. Park, "The stability of nano fumed silica particles and its influence on chemical mechanical planarization for interlayer dielectrics." Japanese Journal of Applied Physics, Part 1: Regular Papers, Short Notes and Review Papers 41(7A): p. 4509-4512, 2002. [44] James, D.B., "Cmp polishing pads." Springer Series in Materials Science 69(Chemical-Mechanical Planarization of Semiconductor Materials): p. 167-213, 2004. [45] Philipossian, A. and S. Olsen, "Effect of pad surface texture and slurry abrasive concentration on tribological and kinetic attributes of ild cmp." Materials Research Society Symposium Proceedings 767(Chemical-Mechanical Planarization): p. 119-125, 2003. [46] Krishnashree, A., "Evaluation a nd characterization of polyurethance cmp polishing pads." 1998, Clarkson University: Postdam, NY. [47] Li, I., K.M. Forsthoefel, K.A. Rich ardson, Y.S. Obeng, W.G. Easter, and A. Maury, "Dynamic mechanical analysis (dma) of cmp pad materials." Materials Research Society Symposium Proceedings 613(Chemical-Mechanical Polishing 2000: Fundamentals and Materials Issues): p. E7.3.1-E7.3.10, 2001. [48] Lu, H., Y. Obeng, and K.A. Richar dson, "Applicability of dynamic mechanical analysis for cmp polyurethane pad studies." Materials Characterization 49(2): p. 177-186, 2002. [49] Lawing, A.S. "Pad conditioning a nd pad surface characterization in oxide chemical mechanical polishing." in Materials Research Society Symposium Proceedings 2002. [50] Hepburn, C., Polyurethane elastomers. London: Applied Science 1982.
PAGE 122
110 [51] Zantye, P., S. Mudhivart hi, A.K. Sikder, A. Kumar, S. Ostapenko, and J. Harmon, "Investigation of mechanical integrity and its effect on polishing for novel polyurethane polishing pad." Materials Research Society Symposium Proceedings 816(Advances in Chemical-Mechani cal Polishing): p. 113-118, 2004. [52] Belyaev, A., F. Diaz, W. Moreno, S. Os tapenko, F. Pachano, I. Tarasov, and D. Totzke, "Non-destructive evaluation of cmp pads using scanning ultrasonic technique." Proceedings Electrochemical Society 2000-26(Chemical Mechanical Planarizatio n IV): p. 94-101, 2001. [53] Mark, L., "Renesas to expand 300-mm fab in 90-, 65-nm push", 2004, http://www.eetimes.com/news/semi/show Article.jhtml;jsessionid=IVTYDZ14NN JYUQSNDBCCKHSCJUMEKJVN?articleID=22101707 August2005, [54] Preston, F.W., "The theory and de sign of plate-glass polishing machines." Journal of the Society of Glass Technology 11: p. 214-56, 1927. [55] Moon, Y., "Mechanical as pects of the material rem oval mechanism in chemical mechanical polishing (cmp)." 1999, Univ ersity of California at Berkeley: Berkeley. [56] Pohl, M.C., Griffiths, D.A., "The impor tance of particle si ze to the performance of abrasive particles in the cmp process." Journal of Electronic Materials 25(10): p. 1612-1616, 1996. [57] Sundararajan, S., D.G. Thakurta, D.W. Schwendeman, S.P. Murarka, and W.N. Gill, "Two-dimensional wafer-scale chemical mechanical planarization models based on lubrication theory and mass transport." Journal of the Electrochemical Society 146(2): p. 761-766, 1999. [58] Thakurta, D.G., C.L. Borst, D.W. Sc hwendeman, R.J. Gutmann, and W.N. Gill, "Three-dimensional chemical mechanical planarization slurry flow model based on lubrication theory." Journal of the Electrochemical Society 148(4): p. G207G214, 2001. [59] Cho, C.H., Park, S.S., Ahn, Y., "Three-dimensional wafer scale hydrodynamic modeling for chemical mechanical polishing." Thin Solid Films 389: p. 254-260, 2001. [60] Subramanian, R.S., Zhang, L., Babu, S.V., "Transport phenomena in chemical mechanical polishing." Journal of The Electrochemical Society 146(11): p. 42634272, 1999. [61] Runnels, S.R., Miceli, F., Kim, I., "Va lidation of a large area three-dimensional erosion simulator for chemical mechanical polishing." Journal of The Electrochemical Society 146(12): p. 4619-4625, 1999.
PAGE 123
111 [62] Borst, C.L., D.G. Thakurta, W.N. Gill, and R.J. Gutmann, "Surface kinetics model for silk chemical mechanical polishing." Journal of the Electrochemical Society 149(2): p. G118-G127, 2002. [63] Yu, T.K., Yu, C. C., and Orlowski, M. "A statistical polishing pad model for chemical-mechanical polishing." in Proceedings of the 1993 IEEE International Electronic Devices Meeting 1994. [64] Zhao, Y.W. and L. Chang, "A micr o-contact and wear model for chemicalmechanical polishing of silicon wafers." Wear 252(3-4): p. 220-226, 2002. [65] Seok, J., C.P. Sukam, A.T. Kim, J.A. Tichy, and T.S. Cale, "Multiscale material removal modeling of chemical mechanical polishing." Wear 254(3-4): p. 307320, 2003. [66] Shi, F.G., Zhao, B., "Modeling of chem ical-mechanical polishing with soft pads." Applied Physics A-Materi als Science and Processing 67(2): p. 249-252, 1998. [67] Zhao, B., Shi, F.G., "Chemical mech anical polishing: Threshold pressure and mechanism." Electrochemical and Solid State Letters 2(3): p. 145-147, 1999. [68] Bastawros, A., A. Chandra, Y. Guo, and B. Yan, "Pad effects on material-removal rate in chemical-mechanical planarization." Journal of Electronic Materials 31(10): p. 1022-1031, 2002. [69] Seok, J., A.T. Kim, C.P. Sukam, A. Ji ndal, J.A. Tichy, R.J. Gutmann, and T.S. Cole, "Inverse analysis of material removal data using a multiscale cmp model." Microelectronic Engineering 70(2-4): p. 478-488, 2003. [70] Larsen-Basse, J., Liang, H., "Probabl e role of abrasion in chemo-mechanical polishing of tungsten." Wear 233-235: p. 647-654, 1999. [71] Trogolo, J.A., Rajan, K., "Near surfac e modification of silica structure induced by chemical/mechanical polishing." Journal of Materials Science 29: p. 4554-4558, 1994. [72] Qin, K., "Multi-scale modeling of the slurry flow and the material removal in chemical mechanical polishing.", in Chemical Engineering 2003, University of Florida: Gainesville. [73] Venkatesh, V.C., S. Izman, and S.C. Mahadevan, "Electro-chemical mechanical polishing of copper and chemical mechanical polishing of glass." Journal of Materials Processing Technology 149(1-3): p. 493-498, 2004. [74] Maury, D., Ouma, D., Boning, D., Chung, J. "A modification to preston's equation and impact on pattern density effect modeling." in Advanced Metallization and Interconnect Sy stems for ULSI Applications 1997.
PAGE 124
112 [75] Wrschka, P., J. Hernandez, Y. Hsu, T. S. Kuan, G.S. Oehrlein, H.J. Sun, D.A. Hansen, J. King, and M.A. Fury, "Po lishing parameter dependencies and surface oxidation of chemical mechanical polishing of al thin films." Journal of the Electrochemical Society 146(7): p. 2689-2696, 1999. [76] Brown, N.J., P.C. Baker, and R.T. Maney, "Optical polishing of metals." Proceedings of SPIE-The International Society for Optical Engineering 306: p. 42-57, 1981. [77] Johnson, K.L., Contact mechanics. Ca mridge: Cambridge University Press 1985. [78] Liu, C.-W., B.-T. Dai, W.-T. Tseng, and C.-F. Yeh, "Modeling of the wear mechanism during chemical-mechanical polishing." Journal of the Electrochemical Society 143(2): p. 716-21, 1996. [79] Runnels, S.R., "Feature-scale flui d-based erosion modeling for chemicalmechanical polishing." Journal of the Electrochemical Society 141(7): p. 1900-4, 1994. [80] Runnels, S.R. and L.M. Eyman, "T ribology analysis of chemical-mechanical polishing." Journal of the Electrochemical Society 141(6): p. 1698-701, 1994. [81] Runnels, S.R. and P. Renteln, "Mode ling the effect of polish pad deformation on wafer surface stress distributions during chemical-mechanical polishing." Proceedings Electrochemical Society 93-25(Interconnects, Contact Metallization, and Multil evel Metallization/Reli ability for Semiconductor Devices, Interconnects, and Thin Insu lator Materials): p. 110-21, 1993. [82] Tseng, W.-T. and Y.-L. Wang, "Re-examination of pressure and speed dependences of removal rate during ch emical-mechanical polishing processes." Journal of the Elect rochemical Society 144(2): p. L15-L17, 1997. [83] Zhang, F. and A. Busnaina, "The role of particle adhesion and surface deformation in chemical mech anical polishing processes." Electrochemical and Solid-State Letters 1(4): p. 184-187, 1998. [84] Zhang, F., A.A. Busnaina, and G. Ah madi, "Particle adhesion and removal in chemical mechanical polishing and post-cmp cleaning." Journal of the Electrochemical Society 146(7): p. 2665-2669, 1999. [85] Ahmadi, G. and X. Xia, "A model for mechanical wear and abrasive particle adhesion during the chemical m echanical polishing process." Journal of the Electrochemical Society 148(3): p. G99-G109, 2001. [86] Mazaheri, A.R. and G. Ahmadi, "A model for effect of colloidal forces on chemical mechanical polishing." Journal of the Electrochemical Society 150(4): p. G233-G239, 2003.
PAGE 125
113 [87] Mazaheri, A.R. and G. Ahmadi, "Modeli ng the effect of bumpy abrasive particles on chemical mechanical polishing." Journal of the Electrochemical Society 149(7): p. G370-G375, 2002. [88] Su, Y.-T., "Investigation of removal rate properties of a floating polishing process." Journal of the Elect rochemical Society 147(6): p. 2290-2296, 2000. [89] Zhang, J.Z., Huang, S. K., Toh, W. S., Tay, W. S., Chen, F., Zhou, B. -B. "Optimization of pad conditioning for stable oxide cmp process." in CMP-MIC Conference Santa Clara 1997. [90] Zhao, B., and Shi, F.G. "Chemical m echanical polishing in ic processes: New fundamental insights." in Proceedings of the chemical-mechanical polish planarization for ULSI multilevel interconnection conference (CMP-MIC) Santa Clara, CA 1999. [91] Luo, J.F., Dornfeld, D. A., "Materia l removal regions in chemical mechanical planarization for submicron integrated circuit fabrication: Coupling effects of slurry chemicals, abrasive size distribution, and wafer-pad contact area." IEEE Transactions on Semiconductor Manufacturing 16(1): p. 45-56, 2003. [92] Luo, J.F., Dornfeld, D. A., "Effects of abrasive size distribution in chemical mechanical planarization: Modeling and verification." IEEE Transactions on Semiconductor Manufacturing 16(3): p. 469-476, 2003. [93] Fu, G.H., Chandra, A.,Guha, S., Subha sh, G., "A plasticity-based model of material removal in chemical-mechanical polishing (cmp)." IEEE Transactions on Semiconductor Manufacturing 14(4): p. 406-417, 2001. [94] Paul, E., "A model of ch emical mechanical polishing." Journal of the Electrochemical Society 148(6): p. G355-G358, 2001. [95] Zhao, Y.W., Chang, L., Kim, S.H., "A mathematical model for chemicalmechanical polishing based on forma tion and removal of weakly bonded molecular species." Wear 254: p. 332-339, 2003. [96] Qin, K., B. Moudgil, and C.-W. Park, "A chemical mechanical polishing model incorporating both the chemical and mechanical effects." Thin Solid Films 446(2): p. 277-286, 2004. [97] Castillo-Mejia, D., S. Gold, V. Bu rrows, and S. Beaudoin, "The effect of interactions between water and polishing pads on chemical mechanical polishing removal rates." Journal of the Electrochemical Society 150(2): p. G76-G82, 2003. [98] Ouma, D.O., D.S. Boning, J.E. Chung, W. G. Easter, V. Saxena, S. Misra, and A. Crevasse, "Characterization and mode ling of oxide chemical-mechanical
PAGE 126
114 polishing using planarization length and pattern density concepts." IEEE Transactions on Semiconductor Manufacturing 15(2): p. 232-244, 2002. [99] Warnock, J., "A 2-dimensional pr ocess model for chemimechanical polish planarization." Journal of the Electrochemical Society 138(8): p. 2398-2402, 1991. [100] Lin, Y.Y. and S.P. Lo, "A study of a finite element model for the chemical mechanical polishing process." International Journal of Advanced Manufacturing Technology 23(9-10): p. 644-650, 2004. [101] Gan, T., T. Tugbawa, B. Lee, D.S. Boning, and S. Jang, "Modeling of reverse tone etchback shallow trench isolation chemical mechanical polishing." Journal of the Electrochemical Society 148(3): p. G159-G165, 2001. [102] Chekina, O.G., Keer, L.M., Liang, H., "Wear-contact problems and modeling of chemical mechanical polishing." Journal of The Electrochemical Society 145(6): p. 2100-2106, 1998. [103] Jeng, Y.R. and H.J. Tsai, "Improved model of wafer/pad powder slurry for cmp." Journal of the Elect rochemical Society 150(6): p. G348-G354, 2003. [104] Thakurta, D.G., D.W. Schwendeman, R. J. Gutmann, S. Shankar, L. Jiang, and W.N. Gill, "Three-dimensional wafer-scale copper chemical-mechanical planarization model." Thin Solid Films 414(1): p. 78-90, 2002. [105] Kao, Y.C., C.C. Yu, and S.H. Sh en, "Robust operation of copper chemical mechanical polishing." Microelectronic Engineering 65(1-2): p. 61-75, 2003. [106] Lai, J.Y., Saka, N., Chun, J.H., "Evolu tion of copper-oxide damascene structures in chemical mechanical polishing I. Contact mechanics modeling." Journal of The Electrochemical Society 149(1): p. G31-G40, 2002. [107] Shaw, D., Chang, J., "A method to improve the efficiency of cmp process." IEEE Transactions on Components And Packaging Technologies 24(4): p. 661-666, 2001. [108] Fu, G.H. and A. Chandra, "A model fo r wafer scale variation of removal rate in chemical mechanical polishing based on elastic pad deformation." Journal of Electronic Materials 30(4): p. 400-408, 2001. [109] Xia, X. and G. Ahmadi, "Surface rem oval rate in chemical-mechanical polishing." Particulate Science and Technology 20(3): p. 187-196, 2002. [110] Paul, E., Brusic, Vlasta., Sun, Fred., Zhang, Jian., Vacassy, Robert., and Kaufman, Frank. "A model of cu-cmp." in Materials Research Society Symposia Proceedings 2004.
PAGE 127
115 [111] Chen, P.H., H.C. Shih, B.W. Huang, and J.W. Hsu, "Catalytic-pad chemical kinetics model of cmp." Electrochemical and Solid State Letters 6(12): p. G140G142, 2003. [112] Komanduri, R., Chandrasekaran, N., Ra ff, L.M., "Effect of tool geometry in nanometric cutting: A molecular dynamics simulation approach." Wear 219: p. 84-97, 1998. [113] Chagarov, E. and J.B. Adams, "Mol ecular dynamics simulations of mechanical deformation of amorphous silicon dioxide during chemical-mechanical polishing." Journal of Applied Physics 94(6): p. 3853-3861, 2003. [114] Ye, Y., Biswas, R., Morris, J.R., Bastaw ros, A., and Chandra, A. "Simulation of nanoscale polishing of copper with molecular dynamics." in Materials Research Society Symposium Proceedings 2002. [115] Ju, S.P., Weng, C.I., Hwang, C. C., "Damascene process simulation using molecular dynamics." Journal of Applied Physics 92(12): p. 7062-7069, 2002. [116] Zhou, C.H., L. Shan, J.R. Hight, S. Danyluk, S.H. Ng, and A.J. Paszkowski, "Influence of colloidal abrasive size on ma terial removal rate and surface finish in sio2 chemical mechanical polishing." Tribology Transactions 45(2): p. 232-238, 2002. [117] Greenwood, J.A. and J.B.P. Williamson, "Contact of nominally flat surfaces." Proceedings of the Royal Society of London, Series A: Mathematical, Physical and Engineering Sciences 295(1442): p. 300-19, 1966. [118] Bouvet, D., P. Beaud, P. Fazan, R. Sanjines, and E. Jacquinot, "Impact of the colloidal silica particle size on physical vapor deposition tungsten removal rate and surface roughness." Journal of Vacuum Science & Technology, B: Microelectronics and Nanometer Structures 20(4): p. 1556-1560, 2002. [119] Gouda, S.D., Bastawros, A., Chandra A.,. "Multi-scale characterization of pad role on material removal rate in cmp." in Materials Research Society Proceedings 2003. [120] Tregub, A., Ng, G., Moinpour, M. "Pad degradation during cmp process: Effect of soak in slurry and water on thermal and mechanical properties of the cmp pads." in Materials Research Society Proceedings 2003. [121] Tregub, A., Moinpour, M., and Sorooshi an, J. "Effect of cmp pad exposure to aqueous media on the pad properties. Part i. Dynamic mechanical and modulated dsc analysis." in Materials Research Society Proceedings 2002. [122] Obeng, Y.S., Forsthoefel, K.M., Richardson, K.A., Burton, R.H. "Mechanistic aspects of the relationship between cmp consumables and polishing characteristics." in Electrochemical Society Proceedings 1998.
PAGE 128
116 [123] Jeng, Y.-R., Huang, Pay-Yau., "A ma terial removal rate model considering interfacial micro-contact w ear behavior for chemical mechanical polishing." Journal of Tribology 127: p. 190-197, 2005. [124] Kim, H.J., H.Y. Kim, H.D. Jeong, E.S. Lee, and Y.J. Shin, "Friction and thermal phenomena in chemical mechanical polishing." Journal of Materials Processing Technology 130: p. 334-338, 2002. [125] Basim, G.B., "Formulation of engin eered particulate systems for chemical mechanical polishing applications", in Materials Science and Engineering 2002, University of Florida: Gainesville. [126] Basim, G.B., Vakarelski, I.U., Moudgil, B.M., "Role of interaction forces in controlling the stability and polishing performance of cmp slurries." Journal of Colloid And Interface Science 263(2): p. 506-515, 2003. [127] Hogan, J.M., C.J. Pearson, T.H. Roge rs, and J.R. White, "Humidity characteristics of hr [high resilient] and hot urethane foams." Journal of Cellular Plastics 9(5): p. 219-25, 1973. [128] Fukushima, D., Y. Tateyama, and H. Yano. "Impact of pad conditioning on cmp removal rate and planarity" Advanced Metallization Conf. Montreal, Canada, p. 699-704, 2002. [129] Oliver, M.R., R.E. Schmidt, and M. Robinson, "Cmp pad surface roughness and cmp removal rate." Proceedings Electrochemical Society 2000-26(Chemical Mechanical Planarizatio n IV): p. 77-83, 2001. [130] Moon, Y., I. Park, and D.A. Dornfeld "Mechanical properties and relationship to process performance of the polishing pad in chemical mechanical polishing (cmp) of silicon" Ph.D Thesis, University of California, Berkeley, 1998. [131] Choi, W., J. Abiade, S.-M. Lee, and R.K. Singh, "Effects of slurry particles on silicon dioxide cmp." Journal of the Electrochemical Society 151(8): p. G512G522, 2004. [132] Lu, Z., S.-H. Lee, S.V. Babu, and E. Matijevic, "The use of monodispersed colloids in the polishing of copper and tantalum." Journal of Colloid and Interface Science 261(1): p. 55-64, 2003. [133] Bielmann, M., U. Mahajan, and R.K. Singh, "Effect of particle size during tungsten chemical mechanical polishing." Electrochemical and Solid-State Letters 2(8): p. 401-403, 1999. [134] Jung, S.-H. and R.K. Singh, "Effect of nano-size silica abrasives in chemical mechanical polishing of copper." Materials Research Society Symposium Proceedings 816(Advances in Chemical-Mechani cal Polishing): p. 49-52, 2004.
PAGE 129
117 [135] Abiade, J.T., S. Yeruva, B. M oudgil, D. Kumar, and R.K. Singh, "Characterization of the chem ical effects of ceria slurri es for chemical mechanical polishing." Materials Research Society Symposium Proceedings 867(ChemicalMechanical Planarization--Integration, Technology and Reliability): p. 69-74, 2005. [136] Abiade, J.T., W. Choi, and R.K. Singh, "Effect of ph on ceria -silica interactions during chemical mechanical polishing." Journal of Materials Research 20(5): p. 1139-1145, 2005. [137] Choi, W., U. Mahajan, S.-M. Lee, J. Abiade, and R.K. Singh, "Effect of slurry ionic salts at dielectric silica cmp." Journal of the Electrochemical Society 151(3): p. G185-G189, 2004. [138] McGrath, J., Davis, C., McGrath, Jim., "The effect of thin film stress levels on cmp polish rate for peteos wafers." Journal of Materials Processing Technology 132: p. 16-20, 2003.
PAGE 130
118 BIOGRAPHICAL SKETCH Suresh Babu Yeruva was born on Feb 18th, 1978, in Machilipatnam, Andhra Pradesh, India. He finished his undergradua te education at the Indian Institute of Technology, Madras (IITM), India. He pursued his Master of Science at Washington State University in the Department of Mate rials Science and Engineering. In 2001, he started his Ph.D. program with Professor Br ij Moudgil at the University of Florida.
|
|