CMP Modeling as Part of Design for Manufacturing
David Dornfeld
Will C. Hall Professor of Engineering
Laboratory for Manufacturing and Sustainability
Department of Mechanical Engineering
University of California
Berkeley CA 94720-1740
http://lmas.berkeley.edu
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
• Modeling objectives and perspective
• CMP process model development
• Short review
• Towards design for manufacturing (DFM)
Outline
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Level IFeature prediction, control, andoptimization in an iterative designand process planning environment
Design: HighManufacturing: HighFinishing: High
Level IIFeature prediction, control, andoptimization through the selection ofa manufacturing plan in an "over-the-wall" design-to-manufacturingenvironment
Design: LowManufacturing: HighFinishing: High -> low
Level IIIFeature prediction and controlthrough limited adjustments to apre-established manufacturingprocess
Design: LowManufacturing: LimitedFinishing: High -> low
Level IVFeature prediction for finishingprocess planning, finishing tooltrajectories and sensor-feedbackstrategies
Design: LowManufacturing: LowFinishing: High
So
ftwa
re d
r ive
n
Ha
rdw
are
dri
ven
Levels of Flexibility - Design to Manufacturing
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Minimum cost/CoOMaximum productionMaximum flexibilityMaximum qualityMinimum environmental & social impactBroadest integration
***
Through software
Modeling Roadmap for maximum impactFunctional
Model
Feedback (validation)
Integrationwith CAD
Feedback (validation)
Feedback (validation)
Include “islands of automation” and existing models)
Include supply chain with
constraints (e.g. “quality gates” )
Prototype based
on model
Feedback (validation)
Extend to “socialimpact” constraints
(green, sustainability,health, safety, etc.)
Feedback (validation)
Feedback (validation)
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Is there need for this?
Design Manf’g
Design Manf’g
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
+ Manf’gDesign
What you see depends on where you are standing!
+DesignManf’g
Source: Y. Granik, Mentor Graphics
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
What’s your world view?
Design
Process
Process
Design
Process
Design
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Mechanical Phenomena
Chemical Phenomena
Interfacial and Colloid
Phenomena
Components of Chemical Mechanical Planarization
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Scale Issues in CMP
From E. Hwang, 2004
Scale/sizenm µm mm
Material Removal
Mechanical particle forcesParticle enhanced chemistry
ChemicalReactions
ActiveAbrasives
Pores,Walls
Grooves
Tool mechanics,Load, Speed
critical features dies
Pad
Mechanism
Layoutwafer
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Bulk Cu CMP Barrier polishing W CMP Oxide CMP Poly-Si CMP
Physical models of material removal mechanism in abrasive scale
Chemical reactions
Bulk Cu slurry Barrier slurry W slurry Oxide slurry Poly-Si slurry
Mechanical material removal mechanism in abrasive scale
Abrasive type, size and concentration
[oxidizer], [complexing agent], [corrosion inhibitor],
pH …
Pad asperity density/shape
Pad mechanical propertiesin abrasive scale
Pad properties in die scale
Slurry supply/ flow patternin wafer scale
Wafer scale pressure NU Models of WIWNU
Models ofWIDNU
Topography
Wafer scale velocity profile
Wafer bending with zone pressures
Better control of WIWNU
Reducing ‘Fang’
Small dishing & erosion
Ultra low-k integration
Smaller WIDNU
Reducing slurry usageUniform pad performance
thru it’s lifetimeLonger pad life time
Reducing scratch defects
Better planarization efficiency
E-CMPPad groove
Pad design
Fabrication
Test
Fabrication technique
Slurry supply/ flow pattern in die scale
Cu CMP
model
design goal
Pad development
PatternMIT model
DornfeldDoyle
An overview of CMP research in Berkeley
Talbot
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
CMP Modeling History in SFR/FLCC*P
rest
on
’s E
qn
.M
RR
= C
PV
Com
bin
ed
eq
n.
R=
CM
/(C
+M
)before SFR/FLCC
Lu
o (
SF
R)
MR
R=
N
Vol
Ch
oi
(FL
CC
) M
RR
=
Tri
path
i (F
LC
C)
Tri
bo-e
lect
ro-c
hem
ical
mod
el
Interfacial/colloidal effects
* According to Dornfeld
DfM/MfD
Computational efficiencyFlexible in scale
Process links
now
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Interactions between Input Variables
Four Interactions: Wafer-Pad Interaction; Pad-Abrasive Interaction;
Wafer-Slurry Chemical Interaction; Wafer-Abrasive Interaction
Polishing pad
Abrasive particles in Fluid (All inactive)
Pad asperity
Active abrasives on Contact area
Vol Chemically Influenced Wafer Surface
Wafer
Abrasive particles on Contact area with number N
Source: J. Luo and D. Dornfeld, IEEE Trans: Semiconductor Manufacturing, 2001
Velocity V
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Pad Materials/Shape Effects
Dishing d
3
Erosion e
2
1
Df
S1=Df1
S=S0
H=Hcu0+Hox0 H= Hstage1
Hcu0
Hox0
Pad/wafer contact modes in damascene polishing
Linear Viscoelastic Pad0 20 40 60 80 100 120 140 160 180 200
0
50
100
150
200
250
300
350
400
450
500
Polishing Time t (second)
Ste
p H
eigh
t S
(nm
)
PDi= 0.1PDi= 0.2PDi= 0.3PDi= 0.4PDi= 0.5PDi= 0.6PDi= 0.7PDi= 0.8PDi= 0.9
Stage 1
Stage 2
Stage 3
Dishing and erosion
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Effect of Pattern Density - Planarization Length (PL)
ILD
Metal lines
Planarization Length
High-density region
Low-density region
Global step
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Effective pattern density
a=320um
a=640um
a=1280um
< Effective density map >
< Test pattern >
< Post CMP film thickness prediction at
die-scale >
Modeling of pattern density effects in CMP
Planarization length (window size) effect on “Up area”
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
PAD
Z(x,y)
Reference height (z=0)
Z_pad
Z(x,y)
Z_padz
dz
padZyxZ
zyxZzPDFdzzPDFdensityasperityKpyxF_),(
0
)),(())()(()_(),(
Feature level interaction between pad asperities and pattern topography
die
dxdyyxFtentF ),(_
F_tent > F_die ? F_tent < F_die ?
++Z_pad --Z_pad
No
Yes
No
Yes
Z_pad
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Asperity Height (µm)
Probability Density (µm-1)
(source : A.Scott Lawing, NCCAVS, CMPUG 5/5/2004)
pD
ab
active asperities
Characterization of Pad Surface
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
asperities# ),(),(
),( yxRdt
yxdzyxMRR a
),( yxz
New model
pattern density effect
pad asperity height distribution
polishing speed
fitting parameter accounting for chemical reactions, abrasive size distribution etc.
pad/film properties
Mean distance between asperities
hardness of material polished
abrasive particle size
Model for the simulation
),( 4/7
2/3
2/3*4/12
** )(),,(),(
),(yxz
zpw
pa
pad
dAHDyxyxPDDH
VERRCyxMRR
asperity radius
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Pattern density Line widthLine space
CMP model
Chip Layout
HDP-CVD Deposition ModelCMP Input Thickness
Evolution Nitride thinning
Modeling Overview
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Removal Rate (RR)
Adding the electro-chemical effects
Slurry chemistry(pH, conc. of oxidizer, inhibitor & complexing agent)
Pad propertieslayers’ hardness, structure
Abrasive Type, size & conc.
Polishing conditions(pressure P, velocity V)
• Develop a transient tribo-electro-chemical model for material removal during copper CMP– Experimentally investigate different components of the model
• Using above model develop a framework for pattern dependency effects.
Polished material
Planarization, Uniformity, Defects
Incoming topography
CMPModel
1. Passivation Kinetics2. Mechanical Properties
of Passive Film3. Abrasive-copper Interaction
Frequency & Force
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Application: Polishing induced stress
Pressure concentrated locally (about 300 psi)
Risk of cracking in the sub layers
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
FEM Analysis: Model
BOUNDARY CONDITIONS:- Fixed at the bottom- Periodic Boundary Conditions (symmetry)
LOADS:- Downward Constant Pressure – 2psi- Horizontal Shear (friction) stress – 0.7psi
LOW-K LayerE = 5 – 20 GPa ; α = 0.25
TANTALUM LayerE = 185.7 GPa ; α = 0.34
COPPER LayerE = 129.8 GPa ; α = 0.34
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
FEM Analysis in CMPVon Mises stresses
Low-k: E = 5GPa
Step1
Step2
Step3
Step3
Low-k: E = 20GPa
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
• Present methods treat CMP process as a black box; are blind to process & consumable parameters
• Need detailed process understanding– For modeling pattern evolution accurately
• Present methods do not predict small feature CMP well– For process design (not based on just trail and error)
• Multiscale analysis needed to capture different phenomena:– At sufficient resolution & speed
• CMP process less rigid than other processes: possibility of optimizing consumable & process parameters based on chip design– MfD & DfM
• Source of pattern dependence is twofold: – Asperity contact area (not addressed yet)– Pad hard layer flexion due to soft layer compression (addressed by previous
models)
Modeling Challenges
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Source: Praesegus Inc.
Extensive test/measurements required
Specific to particular processing conditions
• captures only 1 source of pattern dependency• coarse (resolution ~10µm)
Present Approach (Praesegus/Cadence, Synopsys)
• Helps in dummy fill -- Design improvement but no process optimization• Optimization should be across process & design: - Need to be able to tune all the available control knobs
Model:
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Pattern Related DefectsLow pattern
densityHigh pattern
density
erosion & dishing
residue film
Initial topography
Non-uniform removal
Local planarization
End point
Over polishing
Nominal Pattern density = Area(high features) / (Total Area)
• ρ(x,y) calculated as a convolution of a weighted function (elliptic) over evaluation window.
• Evaluation window size (R) determined empirically.
),(),(
yx
KyxMRR
R
Time step evolution
• MRR(x,y) = material removal rate at (x,y)
• K = Blanket MRR
• ρ(x,y) = effective pattern density at (x,y)
Present Approach
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Need a “GoogleEarth” view of modeling
We are here
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Head
Platen
Pad
do
wn
-
forc
e
slurry supply
rotation of wafer
head
Wafer 4-12”
Copper
Feature
pad asperity
abrasive particles
100nm-10µm
~1
µm
1-10µmPad asperity
Abrasive
Pad/Wafer
Die
Feature/Asperity
Abrasive Contact
CMP phenomena at different scales
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Pattern Evolution Framework
),(),(
yx
KyxMRR
R
Time step evolution Asperity contact area (µm)
Empirically fit, based onpad flexion (scale=mm)
Space Discretization: Data Structure
STI oxide evolution*
0.112μm/0.1681μm
before 40sec CMP
*Choi, Tripathi, Dornfeld & Hansen, “Chip Scale Prediction of Nitride Erosion in High Selectivity STI CMP,” Invited Paper, Proceedings of 11th CMP-MIC, 2006
• Consumables• Polishing Conditions
Material Removal Model
0
0 )( dtttinF
MRR Cu
Small feature prediction problems
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Effects to Capture
• Multiscale Behavior– Material removal operates on different scales and contributes to
the net material removed in the CMP process– Material removal at any location is affected by its position in
different scales– Different models need to be used to capture behavior at different
scales
• Far-field Effects– Most IC manufacturing processes are only dependant on local
features– CMP performance depends on both local as well as far-field
features
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Pad/Wafer (~m)
Die (~cm)
Asperity (~µm)
Feature (45nm-10µm)
Abrasive contact (10nm)
CMP Model Tree• Tree based data structure will encapsulate both wafer features
and pattern evolution at various scales
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Data Structure• Efficient surface representation is required
– Mesh-based representations allow for fast processing, and have been widely used
• Need to capture repeating features– Use tiles/modular units– For “similar” features, use property inheritance
from modular features
• Multiscale analysis– Use multiresolution meshes – allow for
querying in mm/um/nm scales– Also support querying of far-field features along
with local features
Multiresolution meshes will allow for querying in different scales - resolution will be determined by feature scales; tiling will be used
for repeating features.
nm
m
cm
mm
μm
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Data Structure• Model precision vs. Level of Detail
– Identify tradeoffs between speed of analysis and the accuracy of the models used
• Data Structure design motivated by physical considerations– Tree levels ≡ phenomenon scale– object properties ≡ physical
phenomena.
• Inheritance: – Inherit properties from parents at
higher levels of tree and from generic object at that level
Example of property inheritance from parent features and base features applied in CMP process model
Asperity(feature)
CMP Process Model on
Asperity Scale
Pad(parent)
Properties inherited from pad
Specific asperity
properties
Generic Asperity
Properties inherited from generic feature
ac
cu
rac
y
analysis time
Leve
l of D
etai
l
Tradeoffs between LOD, analysis time, and accuracy
Resolve intosmaller features
Resolve intolarger features
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Multiscale Optimization Example
• Address WIDNU at different levels depending on available flexibility:– Change pad hardness (tree level 1)
• Inflexibility: scratch defects, pad supplier
– Dummy fill (chip, array level)• Inflexibility: design restrictions
– Change incoming topography (feature level)
• Inflexibility: deposition process limitation
– Change chemical reactions, abrasive concentration (abrasive level)
Within die non-uniformityNitride Thinning in STI
University of California at Berkeley
Laboratory for Manufacturing and Sustainability, 2007
Thank you for your attention!