In Situ Friction Measurements in Chemical Mechanical Planarization

In Situ Friction Measurements in Chemical Mechanical Planarization

Jim Vlahakis

PhD. Candidate

Tufts University

20 February 2006

1

Introduction

• Experimental setup• Equipment• Data generation• Data analysis• Results & Discussion

– Coefficient of Friction (COF)– Frequency Analysis

• Sources of Error• Final thoughts

Experimental Setup

• Must accommodate our DELIF experiments– transparent wafer– 9:1 water diluted slurry

to avoid polishing– Framework supports

optics

• Process parameters must be modified to account for laboratory scaling– Wafer size = 3” dia.– Default ω = 60rpm– Flow rate ~ 50cc/min

Equipment

Motor – ½ hp Dayton

Wafer – transparent BK7

Table – 136kgs, steel

Platen – 12” diameter

Force table – AMTI

Polisher – Struers RotoPol31

Motor

Platen

RotoPol-31

Wafer

Force Table

Steel Table

Equipment - Issues

• Alignment of polisher and force table

• Mechanical isolation• Support frame• Alignment of wafer

drive belt

• In our setup, Fz, also includes the weight of any fluid in the system

• Platen runout can influence Fz

Equipment - Force Table

• Decomposes the loading into orthogonal components (forces and moments),

• Accuracy– 355 bits/lb in x and y– 710 bits/pound in z

Equipment - Polisher

• Struers RotoPol 31 table top polisher.

• Rests directly on top of the force platform

• Real time measurements of the wafer/pad interaction forces

• Fz – process downforce• Fx, Fy - friction due to

polishing • Custom LabView software

allows us to select a rotation rate from 0 – 100rpm

Equipment - Wafer

• Transparent glass BK7 wafer

• Wafer concavity mates with drive shaft

• Drive plate (red plastic) ensures positive engagement with wafer drive pins

• Decent amount of “play” allows the wafer some freedom of movement

Data Generation

• Custom LabView software controls force table, digital amplifier and I/O settings

• Front panel seems complicated but is pretty straightforward

• Most settings are “set and forget”

Data Analysis

• Data format - 6 columns, tab delimited

• Each column represents one component (Fx, Fy, etc.)

• Sampling rate = 2kHz• Each data run ~ 20sec• Data file sizes up to

tens of megabytes (ie manageable)

• Accuracy Issues– .007N/bit in x and y– .097N/bit in z– Force table/polisher

alignment

Results & Discussion Coefficient of Friction

Ungrooved FX9 pad

Results & Discussion Coefficient of Friction

Circular grooved FX9 pad

Results & Discussion Coefficient of Friction

xy grooved IC1000 pad

Results & Discussion Coefficient of Friction

xy grooved IC1000 pad – low slurry flow rate

Results & Discussion Coefficient of Friction

• For unvented pad– Larger spread in instantaneous COF, ranging from 0.0 to 3.0 – Indicates the lubrication regime is alternating from hydrodynamic

to boundary lubrication– Larger average COF and larger variation in COF

• Higher velocity decreases COF slightly

• For vented pads– Smaller spreads in COF and smaller average COF– Indicates more consistent lubrication regime– Venting seems to moderate the changes in COF

• high Fz-30rpm-IC1000 dataset seems to show some sort of resonance effect

Results & Discussion Frequency Analysis

• Examine the downforce frequency spectrum

• Which frequencies contribute the most

• Can we learn anything about the various polishing parameters based on the frequency signature

Results & Discussion Frequency Analysis

Ungrooved FX9 pad

Results & Discussion Frequency Analysis

Circular grooved FX9 pad

Results & Discussion Frequency Analysis

xy grooved IC1000 pad

Results & Discussion Frequency Analysis

xy grooved IC1000 pad – low slurry flow rate

Results & Discussion Frequency Analysis

• Features at 120Hz/240Hz/360Hz are grounding issues. Must be filtered out in the future.

• Resonant case (highFz-30rpm-IC1000 pad) shows a strong peak at ~190Hz. May be related to pad’s natural frequency

• Which features are important? What scale should we be looking at?

Sources of Error

• Mechanical Issues– Isolation from external inputs– Bearing runout, unbalanced rotating

components

• Electronic Issues– Noise from other equipment– Appropriate sampling rates– Appropriate filtering

Final Thoughts

• What, exactly, do we want to learn?– How to identify failure modes– A polishing end point– Correlate removal rates with COF

• What are the relevant variables? • Which regions of parameter space do we want to explore?• What is the best way to present this data?

• Thanks to– Intel & Cabot for their sponsorship– Our advisors Vin Manno & Chris Rogers– Fellow researchers at U. of Arizona– Howard Stone at Harvard and Gareth McKinley at MIT