Transcript Document

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
Outline
• Modeling objectives and perspective
• CMP process model development
• Short review
• Towards design for manufacturing (DFM)
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Levels of Flexibility - Design to Manufacturing
Level I
Level II
Level III
Level IV
Feature prediction, control, and
optimization in an iterative design
and process planning environment
Feature prediction, control, and
optimization through the selection of
a manufacturing plan in an "overthe-wall" design-to-ma nufa cturing
environment
Feature prediction and control
through limited adjustments to a
pre-established manufacturing
process
Feature prediction for fi nishing
process planning, fi nishing tool
trajectories and sensor-feedback
strategies
 Laboratory for Manufacturing and Sustainability, 2007
Design: High
Manufa cturing: High
Finishing: High
Design: Low
Manufa cturing: High
Finishing: High -> low
Design: Low
Manufa cturing: Limited
Finishing: High -> low
Design: Low
Manufa cturing: Low
Finishing: High
University of California at Berkeley
Modeling Roadmap for maximum impact
Functional
Model
Feedback
(validation)
Prototype
based
on model
Feedback
(validation)
Minimum cost/CoO
Maximum production
Maximum flexibility
Maximum quality
Minimum environmental
& social impact
Broadest integration
*
*
*
Through software
 Laboratory for Manufacturing and Sustainability, 2007
Integration
with CAD
Feedback
(validation)
Include “islands of
automation” and
existing models)
Extend to “social
impact” constraints
(green, sustainability,
health, safety, etc.)
Feedback
(validation)
Feedback
(validation)
Include supply
chain with
constraints (e.g.
“quality gates” )
Feedback
(validation)
University of California at Berkeley
Is there need for this?
Manf’g
Design
Design
 Laboratory for Manufacturing and Sustainability, 2007
Manf’g
University of California at Berkeley
What you see depends on where you are standing!
+
Design
Manf’g
+
Design
Source: Y. Granik, Mentor Graphics
 Laboratory for Manufacturing and Sustainability, 2007
Manf’g
University of California at Berkeley
What’s your world view?
Process
Process
Process
Design
Design
Design
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Components of Chemical Mechanical Planarization
Mechanical
Phenomena
Chemical
Phenomena
 Laboratory for Manufacturing and Sustainability, 2007
Interfacial and
Colloid
Phenomena
University of California at Berkeley
Scale Issues in CMP
Mechanical particle forces
Particle enhanced chemistry
Material
Removal
Active
Abrasives
Pores,
Walls
Tool mechanics,
Load, Speed
Chemical
Reactions
critical features
nm
Pad
Grooves
wafer
dies
µm
From E. Hwang, 2004
 Laboratory for Manufacturing and Sustainability, 2007
Mechanism
mm
Layout
Scale/size
University of California at Berkeley
An overview of CMP research in Berkeley
Cu CMP
Bulk Cu CMP
Barrier polishing
W CMP
Oxide CMP
Poly-Si CMP
Bulk Cu slurry
Barrier slurry
W slurry
Oxide slurry
Poly-Si slurry
Abrasive type,
size and concentration
[oxidizer],
[complexing agent],
[corrosion inhibitor],
pH …
Dornfeld
Talbot
Mechanical material removal mechanism in abrasive scale
Doyle
Chemical reactions
Physical models of material removal mechanism in abrasive scale
Pattern
Topography
Pad properties in die scale
Wafer scale velocity profile
Pad mechanical properties
in abrasive scale
MIT model
Models of
WIDNU
Slurry supply/ flow pattern
in die scale
Wafer scale pressure NU
Pad asperity density/shape
Better planarization efficiency
Better control of WIWNU
Smaller WIDNU
Pad design
Models of
WIWNU
Reducing slurry usage
Uniform pad performance
thru it’s lifetime
Longer pad life time
Wafer bending with zone pressures
Slurry supply/ flow pattern
in wafer scale
Pad groove
model
design goal
Pad development
 Laboratory for Manufacturing and Sustainability, 2007
Small dishing & erosion
Reducing scratch defects
Reducing ‘Fang’
Ultra low-k integration
E-CMP
Fabrication
technique
Fabrication
Test
University of California at Berkeley
CMP Modeling History in SFR/FLCC*
before
now
Preston’s Eqn.
MRR = CPV
Tripathi (FLCC)
Tribo-electro-chemical model
SFR/FLCC
Interfacial/colloidal effects
* According to Dornfeld
 Laboratory for Manufacturing and Sustainability, 2007
DfM/MfD
Computational efficiency
Flexible in scale
Process links
University of California at Berkeley
Interactions between Input Variables
Four Interactions: Wafer-Pad Interaction; Pad-Abrasive Interaction;
Wafer-Slurry Chemical Interaction; Wafer-Abrasive Interaction
Velocity V
Vol
Chemically
Influenced Wafer
Surface
Abrasive particles
in Fluid (All
inactive)
Wafer
Abrasive
particles on
Polishing pad
Pad asperity
Contact
area with
number N
Active abrasives
on Contact area
Source: J. Luo and D. Dornfeld, IEEE Trans: Semiconductor Manufacturing, 2001
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Pad Materials/Shape Effects
500
PDi=
PDi=
PDi=
PDi=
PDi=
PDi=
PDi=
PDi=
PDi=
Stage 1
450
400
Step Height S (nm)
350
Dishing and erosion
300
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Stage 2
250
200
150
Stage 3
100
50
Linear Viscoelastic Pad
S=S0
H=Hcu0+Hox0
0
0
20
40
60
80
100
120
140
Polishing Time t (second)
160
180
200
Df
Hcu0
1
S1=Df1
H=
Hstage1
Hox0
2
Erosion
e
Pad/wafer contact modes in damascene polishing
 Laboratory for Manufacturing and Sustainability, 2007
3
Dishing
d
University of California at Berkeley
Effect of Pattern Density - Planarization Length
(PL)
High-density region
Global step
Low-density region
ILD
Metal lines
Planarization Length
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Modeling of pattern density effects in CMP
Effective pattern density
Planarization length
(window size) effect
on “Up area”
a=320um
< Test pattern >
a=640um
a=1280um
< Post CMP film
thickness prediction at
die-scale >
 Laboratory for Manufacturing and Sustainability, 2007
< Effective density map >
University of California at Berkeley
Feature level interaction between pad asperities and pattern topography
PAD
Z(x,y)
Z_pad
Reference height (z=0)
Z ( x , y )  Z _ pad
F ( x, y)  Kp 
 (asperity_ density)  (PDF( z  dz)  PDF( z)) (Z ( x, y)  z)
0
F _ tent 
 F ( x, y )dxdy
die
No
No
dz
F_tent > F_die ?
Yes
Z(x,y)
++Z_pad
z
F_tent < F_die ?
Yes
--Z_pad
Z_pad
Z_pad
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Characterization of Pad Surface
Dp
Asperity Height (µm)
Probability
Density (µm-1)
a
b
active
asperities
(source : A.Scott Lawing, NCCAVS, CMPUG 5/5/2004)
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Model for the simulation
MRR ( x, y )  
dz ( x, y )
 Ra ( x, y )  # asperities
dt
z ( x, y )
fitting parameter accounting for chemical reactions, abrasive size distribution etc.
abrasive
asperity pad/film polishing
particle size radius properties speed
pad asperity height
distribution
New model
MRR( x, y)  C **
R 2 Ra
1/ 4
3/ 2
Dp
Hw
hardness of
material
polished
 Laboratory for Manufacturing and Sustainability, 2007
E*
3/ 2
Vp

PD( x, y)
z ( x, y )
z pad
 ( x, y,  ) 7 / 4 AHD( )d
Mean distance pattern density
effect
between
asperities
University of California at Berkeley
Modeling Overview
Chip Layout
Pattern density
Line space
HDP-CVD Deposition Model
Line width
CMP Input Thickness
CMP model
Evolution
 Laboratory for Manufacturing and Sustainability, 2007
Nitride thinning
University of California at Berkeley
Adding the electro-chemical effects
• 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.
Slurry chemistry
(pH, conc. of oxidizer, inhibitor
& complexing agent)
Pad properties
layers’ hardness, structure
Abrasive
Type, size & conc.
Polishing conditions
(pressure P, velocity V)
Polished material
Incoming topography
 Laboratory for Manufacturing and Sustainability, 2007
CMP
Model
1. Passivation Kinetics
2. Mechanical Properties
of Passive Film
3. Abrasive-copper Interaction
Frequency & Force
Removal Rate (RR)
Planarization,
Uniformity, Defects
University of California at Berkeley
Application: Polishing induced stress
Pressure
concentrated locally
(about 300 psi)
 Risk of cracking in
the sub layers
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
FEM Analysis: Model
COPPER Layer
E = 129.8 GPa ; α = 0.34
TANTALUM Layer
E = 185.7 GPa ; α = 0.34
LOW-K Layer
E = 5 – 20 GPa ; α = 0.25
BOUNDARY CONDITIONS:
- Fixed at the bottom
- Periodic Boundary Conditions
(symmetry)
 Laboratory for Manufacturing and Sustainability, 2007
LOADS:
- Downward Constant Pressure – 2psi
- Horizontal Shear (friction) stress – 0.7psi
University of California at Berkeley
FEM Analysis in CMP
Von Mises stresses
Step1
Step2
Step3
Step3
Low-k:
E = 5GPa
 Laboratory for Manufacturing and Sustainability, 2007
Low-k:
E = 20GPa
University of California at Berkeley
Modeling Challenges
•
•
•
•
•
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)
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Present Approach (Praesegus/Cadence, Synopsys)
Extensive test/measurements required
Specific to particular processing conditions
Source: Praesegus Inc.
 Laboratory for Manufacturing and Sustainability, 2007
Model: • captures only 1 source of pattern dependency
• coarse (resolution ~10µm)
• 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
University of California at Berkeley
Low pattern
density
Pattern Related Defects
High pattern
density
Present Approach
Initial
topography
• MRR(x,y) = material removal rate at (x,y)
• K = Blanket MRR
• ρ(x,y) = effective pattern density at (x,y)
MRR( x, y ) 
Non-uniform
removal
Local
planarization
K
 ( x, y )
Time step
evolution
residue film
R
End point
erosion & dishing
Over
polishing
Nominal Pattern density = Area(high features) / (Total Area)
 Laboratory for Manufacturing and Sustainability, 2007
• ρ(x,y) calculated as a convolution of a
weighted function (elliptic) over evaluation
window.
• Evaluation window size (R) determined
empirically.
University of California at Berkeley
Need a “GoogleEarth” view of modeling
We are here
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
slurry
supply
rotation
of wafer
head
force
Pad/Wafer
down-
CMP phenomena at different scales
Head
Die
Pad
100nm-10µm
Feature/Asperity
Featur
e
~1µm
Plate
n
Wafer
4-12”
Copper
1-10µm
pad asperity
abrasive
particles
Pad asperity
Abrasive
Abrasive Contact
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Pattern Evolution Framework
• Consumables
• Polishing Conditions
before
40sec CMP
Material Removal Model

M Cu
RR 
i (t0  t )dt
nF 0
0.112μm/0.1681μm
STI oxide evolution*
Small feature prediction problems
MRR( x, y ) 
K
 ( x, y )
Time step
evolution
Space Discretization:
Data Structure
Asperity contact area (µm)
R
*Choi, Tripathi, Dornfeld & Hansen, “Chip Scale Prediction of Nitride Erosion
in High Selectivity STI CMP,” Invited Paper, Proceedings of 11th CMP-MIC, 2006
 Laboratory for Manufacturing and Sustainability, 2007
Empirically fit, based on
pad flexion (scale=mm)
University of California at Berkeley
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
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
CMP Model Tree
• Tree based data structure will encapsulate both wafer features
and pattern evolution at various scales
Pad/Wafer (~m)
Die (~cm)
Asperity (~µm)
Feature (45nm-10µm)
Abrasive contact (10nm)
 Laboratory for Manufacturing and Sustainability, 2007
University of California at Berkeley
Data Structure
• Efficient surface representation is
required
– Mesh-based representations allow for fast
processing, and have been widely used
m
cm
mm
μm
nm
• 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
 Laboratory for Manufacturing and Sustainability, 2007
Multiresolution meshes will allow
for querying in different scales resolution will be determined by
feature scales; tiling will be used
for repeating features.
University of California at Berkeley
Data Structure
accuracy
• Model precision vs. Level of Detail
– Identify tradeoffs between speed
of analysis and the accuracy of
the models used
Resolve into
smaller features
Resolve into
larger features
analysis time
• 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
 Laboratory for Manufacturing and Sustainability, 2007
Tradeoffs between LOD,
analysis time, and accuracy
Pad
(parent)
Properties
inherited
from pad
Asperity
(feature)
Specific
asperity
properties
Generic
Asperity
Properties
inherited from
generic feature
CMP Process
Model on
Asperity Scale
Example of property inheritance from parent features
and base features applied in CMP process model
University of California at Berkeley
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)
 Laboratory for Manufacturing and Sustainability, 2007
Within die non-uniformity
Nitride Thinning in STI
University of California at Berkeley
Thank you for your attention!
 Laboratory for Manufacturing and Sustainability, 2007