MEASUREMENT OF PAD COMPRESSION DURING CHEMICAL MECHANICAL POLISHING A Thesis submitted by Caprice Gray IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN MECHANICAL ENGINEERING TUFTS UNIVERSITY MEDFORD, MA 02155 AUGUST 2005 ADVISOR: CHRIS ROGERS, SC.D
126
Embed
MEASUREMENT OF PAD COMPRESSION DURING CHEMICAL MECHANICAL POLISHING … · measurement of pad compression during chemical mechanical polishing a thesis submitted by caprice gray in
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
MEASUREMENT OF PAD COMPRESSIONDURING CHEMICAL MECHANICAL
POLISHING
A Thesis submitted by
Caprice Gray
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE
OF MASTER OF SCIENCE IN MECHANICAL ENGINEERING
TUFTS UNIVERSITYMEDFORD, MA 02155
AUGUST 2005
ADVISOR: CHRIS ROGERS, SC.D
Abstract
In the 1980s, the semiconductor industries started to use Chemical Mechanical Pol-
ishing (CMP) to planarize silicon wafers before patterning integrated circuits (ICs).
At the time, little was known about the physics and chemistry involved in CMP. The
push towards miniaturization in the past few decades has led to a desire for a better
understanding of the CMP mechanism. Many research groups have explored CMP
via modeling and experimentation. Experimentalists have difficulty verifying the
results produced by the modeling because it is difficult to “see” what is happening
under the wafer while polishing. In the mid 1990s, Tufts University in conjunc-
tion with the University of Arizona, Intel Corporation and Cabot Microelectronics
Corporation, began an effort to make in-situ measurements of physical properties
underneath the wafer using a technique called Dual Emission Laser Induced Fluo-
rescence (DELIF).
In this work, the DELIF setup at Tufts has been modified such that we can at-
tain instantaneous images of the slurry layer thickness at high spatial and temporal
resolution during polishing. It is now possible to make measurements at an asperity
scale resolution. In this thesis, I present both qualitative and quantitative obser-
vations of the slurry layer as it passes between the polishing pad a wafer features.
ii
Images were analyzed to determine pad roughness, local pad deformation and as-
perity compression during the polishing of wafers with etched steps. By calculating
a relative calibration factor, it is possible to estimate asperity compression during
polishing. The ability to resolve wafer features is limited by pad topography. This
calibration factor allows us to examine slurry film roughness, slurry layer thickness
changes near steps in the wafer, and asperity compression and expansion under the
wafer. Step steepness and heights can be characterized by fitting the average cross
section across a step to a hyperbolic tangent curve. Global pad topography results
in widely varying fluid displacement due to applied down-forces to the wafer over
small areas (6 mm2). Analysis of asperity compression in static images after apply-
ing 70kPa down-force shows that average asperity compression increases with the
reduction of the average slurry layer thickness. Dynamic measurements show that
asperity expansion increases with well depth.
iii
Acknowlegements
This work would not have been possible without help from many sources. First, I
would like thank our funders, Cabot Microelectronics Corporation and Intel Corpo-
ration. The representatives from these companies include Mansour Moinpour and
Chris Barns from Intel, and Sriram Anjur from Cabot. These people have provided
both excellent intellectual input as well as supplies for our experimental research.
Second, I would also like to thank Professor Ara Philipossian and the students of his
lab at the University of Arizona, Tucson, for their continuing input on this project.
I have many people to thank at Tufts University. First and foremost, I would
like to thank Professors Chris Rogers and Vincent Manno for giving a Chemist the
opportunity to learn a bit of engineering. Second, I would like to thank my partner
in this research, Daniel Apone. Dan has been one of the best research partners I have
ever had and this project definitely would not have been possible without his help. I
would also like to acknowledge the other student working on Tufts CMP while I have
been here, Scott Taylor and Jim Vlahakis for their ideas and input. I would like to
thank two professors from outside the Mechanical Engineering Department, Professor
David Wilbur of the Chemistry Department and Professor Bruce Boghosian in the
iv
Mathematics Department. Professor Wilbur helped me with all the spectroscopy
and Professor Boghosian helped with understanding the image statistics. Finally, I
would like to thank everyone who has been a part of TUFTL for the past couple of
years for making it a wonderful place to work.
Last, but not least, I would like to thank my family for their continuing support
throughout my education. Mom and Dad always made it a priority to put my brother
and I through school. We made it, thanks. Nicky, your an awesome brother, but you
didn’t hear that from me. Thanks to all of my grandparents for the love and support
and snail mail, Peggy, Pete, Rene, 9E - you all are some of the most amazing people
Calibration of pixel intensity to the height of a column of fluid is extraordinarily
important for acquiring accurate quantitative results. Unfortunately, calibration for
the modified DELIF procedure described in section 2.2.2 is not straight forward.
Three calibrations techniques are mentioned in this chapter. The calibration tech-
nique that is used for quantitative analysis is described in section 3.2.2. In section
3.3 I describe the problems associated with the calibration technique.
Images are acquired using 4 different wafers, a flat wafer, a wafer with 27 µm
deep wells, a wafer with 14.5 µm deep wells, and a wafer with 5 µm deep wells. A
schematic of the wafers with wells can be seen in figure 3.1. The wafers with 27 and
14.5 µm deep wells serve as both a tool for studying pad deformation (chapter 4)
and a method for calculating the correlation factor between slurry layer thickness
and intensity (section 3.3.2).
36
3.1 Wafer Etching
The 3” diameter, 0.5” thick BK7 optical glass blanks were sent to Intel where the
following etch procedure was performed:
1. Clean glass surface (detergent, DI water rinse, Marangoni dry, 5 minutes oxygen
plasma)
2. Sputter 150.0 nm Au on 5.0 nm Cr
3. Dip coat photoresist
4. Cure photoresist at 80 degrees C for 10 minutes
5. Photopattern resist (expose for 45 s, develop for 120 s)
6. Hand paint wafer edges (not covered by pattern mask), develop, dry and bake
at 80 degrees C for 10 minutes.
7. Wet etch Au and Cr in well regions for 30 s
8. Etch glass in well regions using 40 mL 48 % HF in 360 mL DI water
9. Strip photoresist with acetone and isopropanol
10. Etch the remainder of the Au and Cr
The HF etch time for the 27 µm well was 33 minutes and the etch time for
the 14.5 µm well was 17 minutes. This etching procedure was complicated by the
fact that the wafers had already been mounted onto a gimbal making it hard to
37
deposit the Au/Cr mask and the photoresist. As a result, the wafers displayed
large irregular undercuts with a high density of pinholes. While these undercuts and
pinholes would be unacceptable for state of the art ICs, they are undetectable at the
imaging resolution discussed in this experiment. Step heights were verified using a
profilometer. All well depths were the same height with ±0.01 µm.
The patterning mask covers a square region 2”x2” on the wafer. Figure 3.1 shows
resulting pattern on the wafer after the etch procedure. There are 17 rows of wells.
The top row contains 0.5mm x 0.5mm square wells separated by a distance of 1mm.
The second row contains equally spaced squares slightly smaller than 1mm on each
side such that 1 cm contains the equivalent of 6 squares and 5 spaces. The third
row contains 1mm x 1mm square wells separated by 1mm. The fourth row contains
1.5mm x 1.5mm squares separated by 1.5mm spaces. The fifth row contains 2mm
x 2mm squares separated by 2mm spaces. Each row is separated by 2mm and the
set of five rows repeats across the area. The different well areas and spaces between
them are used to test how fast pad asperities rebound into the wells.
3.2 Calibration
The x and y directions on the images were calibrated by the photographing a mil-
limeter ruler as shown in figure 3.2. The number of pixels between the start of one
hash mark and the next were counted. The resulting image resolution at the present
zoom factor is 4.3 µm per pixel.
Correlation of the intensity values, I, to fluid film thickness, T, recorded in the z
38
Figure 3.1: Wafer etch pattern and well cross section. Note: the cross section is notto scale.
39
Figure 3.2: Image used for x-y pixel calibration
direction is not a trivial task and is therefore the subject of the following subsections.
Previous work has proven that the relationship between fluid film thickness and ratio
is close to linear [16]. Therefore we should be able to express thickness as a function
of intensity, δ(I), as shown in the following equation:
δ(I) = XI + δ(0), (3.1)
where X is the calibration factor.
3.2.1 Previous Calibration Attempts
A schematic of other possible calibration procedures is shown in figure 3.3. Figure
3.3a is the method that has been used in previous research [33] in which the slurry fills
40
the gap between the microscopes calibration slides. This method became difficult
to use when we changed the DELIF procedure to use the pad fluorescence as the
excitation source. It became hard to determine how much light from the pad was
able to make it through the microscope slide to excite the Calcein in the slurry.
We determined that the slurry should be in direct contact with the pad for the
calibration. Figure 3.3b shows another calibration procedure we had considered.
One edge of a flat BK7 glass wafer was propped up 1mm from the polishing pad
surface such that there was a known sloping volume of slurry under the wafer. This
technique was not used for a number of reasons. First, it was difficult to determine the
exact slope of the wafer because it was difficult to repeatedly place the microscope
slide in the same place. Second, a single cross section showed a large amount of
asperity noise. Finally, it was difficult to image the edge of the wafer or find the
exact position of the image on the wafer with the decreased field of view in these
experiments. The calibration technique that was finally used is shown in figure 3.4
and is discussed in section 3.2.2.
3.2.2 Relative Thickness
Most of the analysis presented, such as surface roughness, asperity compression, a
fluid layer step heights, are determined by directly comparing one region of the image
to another region or the mean. Therefore, the offset value, δ(0) in equation 3.1, is
canceled in a subtraction. The calibration equation simplifies to
∆δ(I) = X∆I. (3.2)
41
Figure 3.3: Previously used calibration techniques (a) Microscope slides with a shim,(b) Flat wafer over the pad at an angle produced by microscope slide.
The two wafers with the square wells of known depth can be used to determine
X. This calibration is based on the assumption that both the 14.5 µm and 27 µm
deep wells are deep enough such that the tops of the asperities, on average, are not
in contact with the bottoms of the wells, and there is no significant angle between
the wafer and the pad in the region of interest. Since the polishing pad rebounds a
certain unknown distance into each wells, it is not possible to simply say that the
average intensity under a well is equal to the height of the well. However, it may be
possible to assume that the difference in intensities between the 14.5 µm and the 27
µm deep wells is roughly equal to the difference in step heights, 12.5 µm. This leads
to the following equation for calculating X :
X =27− 14.5
I27 − I14.5
. (3.3)
For most runs, the calibration factor attained was 70 µm per ratio unit. A schematic
42
of this calibration procedure can be seen in figure 3.4. This assumes that the wafer
is flat and that the asperity rebound into the wells is independent of well depth.
3.2.3 Absolute Thickness
An absolute calibration would allow the direct assignment of an accurate slurry layer
thickness value at every pixel in an image. If the intensity in the Calcein camera
is zero when no Calcein is present, then the thickness value at zero intensity would
be zero, δ(0) = 0. Additionally, the relative calibration discussed above would be a
valid absolute calibration. However, the intensity values of a ratiod image with no
Calcein are not zero even with dark noise values subtracted from the image because
the image filters can not filter out 100% of the pad fluorescence. The dark-noise
is the average intensity over an image taken with no ambient room lighting and no
laser shot. Since ratio values actually drift from run to run and image to image, it is
extraordinarily difficult to determine δ(0). The lowest ratio intensity values observed
range from 0.3 to 0.5. Techniques for determining the zero point intensity is still a
limitation of the system and are discussed elsewhere [45].
Figure 3.4: Schematic of the method used to determine the relative thickness cali-bration factor.
43
3.3 Experimental Sources of Error
There are two major sources of error in these experiments. First, the average in-
tensity of an image increases with the number of laser shots to that location on the
polishing pad. As a result, the image to image calibration factor, X, increases in time
according to equation 3.2. Possible reasons for this are examined by evaluating the
individual fluorescing components in the DELIF system as observed through each
camera. The second source of error is the uncertainty in determining the ratio value
that corresponds to zero thickness. This is only briefly discussed here because it is
discussed in greater detail elsewhere [45].
3.3.1 DELIF
One of the largest sources of error in this set of experiments is ratio drift in time. The
average image intensity over a run even as short as 20 image shows a positive slope.
The polyurethane pad is known to photodegrade with exposure time to UV light.
However, if the pad fluorescence intensity decreases and the Calcein fluorescence is
dependent on the intensity of the pad, the Calcein fluorescence should also decrease
with pad fluorescence. The ratio should therefore be constant. Unfortunately, the
ratio does increase with time. To investigate ratio drift, each DELIF component
was examined separately. The results are shown in figures 3.5, 3.6, and 3.7. When
examining these figures, it is important to keep in mind that the dark noise values
for both cameras has an intensity of roughly 180. Below is a list of the 5 components
presented in the graphs:
44
1. diamonds = the ”control pad”, which is a sheet of black rubber that we assume
does not fluoresce
2. squares = Calcein in the slurry on top of the control pad
3. triangles = the Fruedenberg FX9 pad
4. circles = the ”run conditions”, which is Calcein in the slurry on top of a
Fruedenberg FX9 pad
5. stars = slurry without Calcein on a Fruedenberg FX9 pad
Figures 3.5 and 3.6 show the intensity data from the above samples as seen
through the Calcein camera and pad camera, respectively. The control pad (dia-
monds) shows intensities that are roughly the same value for both cameras with zero
slope in time. Note that the control pad intensity is above the dark-noise. This is
because the black rubber is not a ”perfect” optical black, meaning that some of the
laser light is reflected back into the cameras. The filters do not filter out 100% of
the laser light, otherwise the control pad line would lie on top of the dark-noise line.
Figure 3.5 examines the intensity as seen through the Calcein camera. Just
above the control pad line are the pad (triangles) and undyed slurry (stars) lines,
which lie roughly on top of each other with a very small positive slope. The fact
that these two lines are not directly on top of the control pad line indicates that
the filters for the Calcein camera don’t completely filter out all of the light from
the pad fluorescence. However, the amount of pad fluorescence is small and roughly
constant and should therefore be accounted for in the calibration factor (provided we
45
can attain an accurate calibration factor). The slope of these lines over this sample
space is negligible. However, the fact that it is positive is problematic since there is
no Calcein present in either of these cases and the pad is know to photodegrade in
time. In the future it may be worth the time investment to observe the pad as seen
through the Calcein camera over a larger number of images to determine if this is a
true slope or experimental noise.
The Calcein line (squares) in figure 3.5 has an average intensity 2.5 times the dark
noise and 2 times greater than the control pad. This means that the Calcein receives
a small excitation from the laser assuming that the control pad does not fluoresce.
Figure 2.9 also shows that the absorbance of Calcein in the region of the laser is not
zero, but it is near a minimum. There is also a positive slope to this line, which also
supports the idea that the Calcein may be an excitation source for itself. Note that
there is a spectral overlap between the Calcein absorbance and emission in figure 2.9
making it possible for the Calcein emission to produce subsequent self-reabsorption.
The average intensity of the Calcein under full run conditions (circles) is only about
three times greater than the Calcein by itself. As shown in figure 2.9, the spectral
overlap between the Calcein absorbance and the pad fluorescence is not very large.
However it is large enough to induce sufficiently detectable Calcein fluorescence with
the current filter setup. In addition to the run condition intensity being 3 times
greater than the Calcein alone, it also seems that the slope of the run conditions
line is about 3 times greater than the Calcein line. This suggests that when the pad
induced Calcein fluorescence is greater, there is more reabsorption of the Calcein
fluorescence by the Calcein.
46
Figure 3.5: DELIF components as seen through the Calcein camera.
47
Figure 3.6 shows intensities of the components as seen through the pad camera,
the denominator of the ratio DELIF images. The control pad (diamonds) and the
Calcein (squares) lines have no slope in time and have a value of a little less than
twice the dark noise. This means that the pad camera is effectively filtering out
all the Calcein fluorescence, but some of the laser light is seeping through the filter
making the total intensity greater than the dark noise. The pad (triangles) and the
undyed slurry on the pad (stars) lines in figure 3.6 have the highest intensities and
both have negative slopes. The negative slopes mean that the pad is photodegrading
in time. It is unclear why the intensity of the pad with undyed slurry over it has a
higher intensity than the pad without slurry. Perhaps different sections of the pad
fluoresce less intensely, or the laser power may have shifted slightly between tests, or
perhaps the slurry particles are reflecting some of the incident light back towards the
cameras. The run condition (circles) intensity is lower the pad and undyed slurry
lines because some of the pad fluorescence is being absorbed by the Calcein. Unlike
the pad and the undyed slurry lines, the run conditions line initially has a negative
slope showing the photodegration of the pad, but the line levels by about image 10.
From examination of these data, one might conclude that the Calcein is reaching
a saturation point in the amount of pad fluorescence it can absorb and perhaps a
higher concentration of Calcein is needed. Increasing the concentration of Calcein is
difficult because of its limited solubility in water.
When the Calcein camera images are divided by the pad camera images, the
result is figure 3.7. Here, the FX9 pad (triangles) and the undyed slurry (stars)
show the smallest ratio values and are very nearly consistent with each other. This
48
Figure 3.6: DELIF components as seen through the pad camera.
49
is because there is no Calcein present in these systems and there is a large amount
of pad fluorescence. There is a slightly positive slope due to the photodegradation
of the pad. The dyed slurry (squares) has the highest intensity because there is no
pad fluorescence. The Calcein is minimally excited by the laser, then it is excited by
itself creating a positively sloping line. The control pad (diamonds) has a constant
ratio in time of approximately 0.6, which is the ratio of the laser light reflected off the
black rubber pad that is not filtered by the cameras. From the control pad line on
this graph, we can conclude the Calcein camera filters are 40 percent more efficient
in filtering the laser light than the pad camera filters. Finally, the line with circles
indicates the observed ratio drift under full run conditions. It has the most positive
slope on this graph because Calcein fluorescence in the numerator has a consistently
positive slope, and the pad fluorescence in the denominator has a negative slope until
image 10 and then levels off.
The effect of Calcein self excitation would be minimized if the spectral overlap
between the pad emission and the Calcein absorption was significantly greater than
the spectral overlap between the Calcein emission and the Calcein absorption. That
overlap can be calculated from figure 2.9 by summing the areas under the overlap
regions of the curves. The pad emission/Calcein absorption overlap is only 2.3 times
greater than the Calcein emission/Calcein Absorption overlap.
3.3.2 Calibration
The next chapter contains both a qualitative and quantitative analysis of the im-
ages. The calibration factor of 70 µm/ratio unit is used for most of the quantitative
50
Figure 3.7: Final ratio of the Calcein camera images divided by the pad cameraimages.
51
analysis. However this calibration factor is only accurate to about ±10 µm because
of the ratio drift discussed in section 3.3.1. This error was calculated using the
method discussed in section 3.2.2 and equation 3.3, using the 27 µm and 14.5 µm
deep wafers. A run of 50 images was taken first with the 14.5 µm deep wafer, then
a run of 50 images of the 27 µm deep wafer, followed by a third run which was a
re-run of the 14.5 µm deep wafer. The calibration factor calculated using the first
and the second run was approximately 60 µm/ratio unit, and the calibration factor
for the second and third run was approximately 80 µm/ratio unit. Calibrations have
been performed for every dynamic experiment involving wafers with square wells. All
values for the calibration factor have been roughly within error of 70±10 µm/ratio
unit. This error may seem large, but keep in mind that the average peak to valley
height for an asperity on a Fruedenberg FX9 polishing pad is approximately 10 µm,
which corresponds to an approximate intensity difference of 0.14 ratio units. The
calibrated height of 0.14 ratio units would be 10±1.4 µm (a 14% error).
The reader should also keep in mind that some graphs presented in chapter 4 (for
example figure 4.27) will show absolute thickness values. These values for absolute
thickness were calibrated using the relative calibration factor discussed in section
3.2.2 and it is still unclear what value of the ratio actually corresponds to zero thick-
ness, the image was calibrated with the relative calibration factor. The difference
between y values will be accurate, however the actual y-values may vary by as much
as 20 µm.
52
Chapter 4
Image Analysis
Thousands of images were taken over the course of this research. A physical inter-
pretation of the images is presented in this chapter. The images are analyzed both
qualitatively and quantitatively. In the qualitative section, the resolution of wafer
and pad features is discussed, and air entrapment between the wafer and the pad is
examined. In the quantitative section, the images are calibrated using the factor dis-
cussed in section 3.3.2. Only the Fruedenberg FX9 pad (see section 4.1.1) was used
to make quantitative measurements for result consistency. First, pad topography is
discussed. Then the slurry layer profile is used to interpret pad compliance to the
wafer features shown in figure 3.1. Three analysis techniques are discussed: calcu-
lating surface roughness, fitting averages of cross section to a step-like function, and
comparing histograms of images to determine the amount of compression at different
pressures.
Some of the images discussed in the following sections are static, and others
53
are dynamic. For the static images, tagged slurry was deposited onto the polishing
pad and the wafer was gently placed on top of the slurry with the force of the
shaft pressing down on it. There was no pad-wafer rotation or conditioning during
the image acquisition. The pad for the static images were conditioned before image
acquisition. The dynamic images were taken with the pad and the wafer both rotating
at 30 RPM, which means the pad-wafer relative velocity at the leading edge of the
wafer was 0.34 m/s. Slurry was injected at the middle of the pad at 50 cc/min. The
pad was conditioned for 20-30 minutes before image acquisition with undyed slurry
and conditioned continuously during acquisition with dyed slurry. Between 50 and
200 images were taken during each dynamic run.
4.1 Qualitative Image Analysis
All images presented in this section are processed images of the ratio obtained by
dividing the Calcein fluorescence intensity by the pad fluorescence intensity. Images
were taken while the platen and the wafer were both in rotation, and those presented
here were picked from thousands of available images. The dark areas in the image
correspond to low values of the ratio of intensities and indicate where the pad and
the wafer are close to each other, or the tops of the pad asperities. Conversely, the
lighter sections of the image indicate high values of the ratio and slurry filled valleys
between pad asperities. The images presented in this section were chosen out of the
thousands taken because they are representative of the various types of pads, wafers
and slurry characteristics observed during the various runs.
54
4.1.1 Images of Different Types of Pads
In this section, DELIF Images of several types of polishing pads will be discussed.
Multiple types of pads were tested such that we could test the resolution of the
DELIF imaging at the new zoom factor. The pads types used are listed below:
• Flat Fruedenberg FX9 (dyed), figure 4.2a
• Flat Fruedenberg FX9 (undyed), figure 4.2b
• k-grooved Fruedenberg FX9, figure 4.3a
• k-grooved Rodel IC1000, figure 4.3b
• Thin-grooved Experimental CMC M2, figure 4.4a
• Flat Experimental CMC M2, figure 4.4b
• xy-grooved Experimental CMC M3, figure 4.4c
The grooving is defined in figure 4.1. The xy-grooves and thin-grooves differ only in
trench depth. The k-grooved pads are grooved in concentric circles and have trenches
shaped like the letter ”k”.
For most pads, images were taken with three different wafer surfaces: a flat wafer,
the wafer with 14.5 µm deep wells, and the wafer with 27 µm deep wells. Figure
4.2a is an image of a dyed flat Fruedenberg FX9 pad, figure 4.2b is an undyed
Fruedenberg FX9 pad. The Fruedenberg FX9 pads are very similar to the industrial
standard Rodel IC1000 pads. The undyed FX9 pad is not smoother than the dyed
pad even though it might appear that way. The focus on the cameras during the
55
Figure 4.1: Pad Grooving (not to scale). The left is an image of an xy-grooved andthin-grooved pad. The right is an image of a k-grooved pad.
run with the undyed FX9 pads was not as sharp as the focus during the dyed pad
run. We did not rerun these images because most of the quantitative measurements
were done on the dyed FX9 pads. The point of taking images on the undyed pads
was to prove that it could be done. The surface roughness measurements (discussed
in section 4.2.2) show the same results for the focused and less focused images.
Figure 4.3 presents the two k-grooved pads. Image ”a” is the Fruedenberg pad and
image ”b” is the Rodel pad. The groove width is 4 times wider on the Fruedenberg
pad than the Rodel pad. In both cases, the grooves contain so much fluid that the
intensity inside the grooves is flooded. There are bands of lower intensities on each
side of the grooves. These bands are not believed to be real thickness, but artifacts
of light bleeding into pixels adjacent to the high intensity grooved regions.
the images also contain good definition of the pad asperities both inside and outside
the wells. The 5 µm deep wells are near the limit of this technique’s ability to resolve
wafer feature over this pad. All of the quantitative analysis is done on these flat
Fruedenberg FX9 pads because of the high DELIF resolution capabilities.
Figures 4.7 and 4.8 show the wells over a k-grooved Rodel IC1000 pad and a
k-grooved Fruedenberg FX9 pad, respectively. It is more difficult to resolve the
well in these images without image enhancement because the amount of slurry in
the groove is significantly greater than outside the groove. The resulting intensity
contrast between regions inside the well and outside the well is greatly diminished.
Therefore resolving any pad features on the asperity level is nearly impossible for this
optical setup. In order to resolve the asperities, we would have to increase the gain
60
Figure 4.5: DELIF image of wafers with etched square wells over the flat, dyedFruedenberg FX9 pad (a) 1 mm2, 14.5 µm deep well (b) 1 mm2, 27 µm deep well
Figure 4.6: DELIF image of a wafer with 5 µm deep wells on an undyed, flat Frue-denberg FX9 with sub-pad. (a) original image (b) enhanced image
61
and/or the exposure time such the contrast between the asperity hills and valleys. At
the new camera settings, the camera pixels in the grooved regions would flood and
the light would bleed into nearby pixels. Therefore, we did not attempt to resolve
the asperities in these images. Enhanced images where the wells are resolved can be
seen in figure 4.9. The enhancement was performed in a similar manner as in figure
4.6b.
Wells over the experimental CMC pads are not shown because these images look
exactly the same as the images of a flat wafer over these pads. The surface roughness
of these pads is an order of magnitude larger than the Fruedenberg FX9 and Rodel
IC1000 pads. Wells could not even be resolved on CMC experimental M2 pads
without grooves. Therefore, the intensity contrast between asperities in the Cabot
pads overshadows even a 27 µm deep well. These large asperities limit our ability to
make any quantitative measurements.
4.1.3 Conditioner Marks on Pads
Most images from the flat Fruedenberg FX9 pads show patterns in the asperities as
indicated by the dashed lines in Figure 4.10. The striation lines in figure 4.10a run
from the top left to bottom right at almost 45 degrees. In figure 4.10b the striation
lines are just off the horizontal with a slightly positive slope. These lines are most
likely gouges in the pad caused by the diamond grit conditioner. Figure 4.10b shows
a slurry layer beneath a 14.5 µm deep, 1000 µm square well etched into the BK7
wafer over a Freudenberg FX9 polishing pad.
62
Figure 4.7: Well resolution over the Rodel IC1000 k-grooved pad (a) 14.5 µm deepwells (b) 27 µm deep wells
63
Figure 4.8: Well resolution over the Fruedenberg FX9 k-grooved pad (a) 14.5 µmdeep wells (b) 27 µm deep wells
Figure 4.9: Enhanced images of figures 4.7 and 4.8.
64
Figure 4.10: Pad striations caused by the conditioner on a flat Fruedenberg FX9polishing pad under (a) a flat wafer (b) 14.5 µm deep wells
65
4.1.4 Air Pockets
Another interesting feature we observed was the formation of air pockets underneath
the 27 µm deep wells. The presence of air pockets during polishing is not typically
considered in fluid flow modeling between the pad and the wafer. Air pockets could
have a significant effect on removal rate near wafer features and post polishing wafer
topography. Figure 4.11 contains images of these air pockets under the 27 µm deep
on a Freudenberg FX9 pad. The arrows in the figure show the direction of slurry
flow. Air pockets were not observed under the 14.5 µm wells and were present in
approximately 50% of the images of the 27 µm deep wells. The majority of the
images of the full square wells showed air bubbles of different sizes moving across
the wells. Figure 4.11a shows an air pocket in the center of a well. The irregular
shape can probably be attributed to slurry flowing around different size asperities.
The majority of the air pockets from the runs with a 27 µm deep wells are similar
to Fig. 4.11b, showing air pockets in the upper left corner of the wells, which is the
trailing edge of the slurry flow. Figure 4.11c shows one of the larger air pockets as
it is dissipating through the contact region just behind the trailing edge of the well.
One can also see the slurry in the pad below the air pocket, implying that one can
estimate the dimensions and volume of the air bubble by comparing the fluorescence
under the well outside of the air bubble as compared to inside the air bubble.
The presence of air pockets is not limited to flat pads. Air can also be seen both
under the wells and in the grooves of grooved pads. Figure 4.12 shows images of
a BK7 wafer with 27 µm wells over a Rodel IC1000 k-grooved pad (a) and over a
Fruedenberg FX9 k-grooved pad. While the grooved in the Fruedenberg pad are
66
Figure 4.11: Air pockets under 27 µm deep wells on a Fruedenberg FX9 pad (a)Middle of well (b) Trailing edge of well (c) Air dissipating from inside to outside thewell
67
much wider than the grooves on the Rodel pad, they both seem to channel the air
pockets underneath the wafer. Most images under on the Rodel pad indicated that
the grooves were partially filled such as the top groove on figure 4.12a. Fewer air
pockets were observed on the Fruedenberg pad, but the pockets observed were larger.
In these images, air pockets beneath the wells were confined to the grooves in the
pad and did not accumulate at the trailing edge of the well as observed on the flat
pad Fig. 4.11.
4.1.5 Pitch Resolution
In order to test the ability to attain a calibration factor for each individual image,
static images were taken of a wafer that had multiple step heights. Figure 4.13 is a
photograph of a microscope image of these steps. The wafer was etched in two steps.
The mask used was an array of linear trenches. The width of the trenches (22 µm)
were about an order of magnitude smaller than the widths of the square wells. As a
result we were able to test the limits of our pitch resolution. The first set of trenches
was etched to a depth (D in figure 4.13) of the 12.5 µm trenches. The mask was
then patterned perpendicular to the first mask, then the second set of trenches was
etched to a depth of 6.5 µm. The resulting depth of where the trenches cross each
other is 19 µm.
The resulting DELIF image is shown on the three dimensional graph in figure
4.14. The thickest fluid layer where the trenches cross are indicated by the array of
dark gray peaks in the image. These peaks vary in height because the pad asperities
are sticking up more into some wells than others. In the upper right corner of this
68
Figure 4.12: The air pockets are confined to the grooves under the 27 µm deep wells(a) Rodel IC1000 k-grooved pad (b) Fruedenberg FX9 k-grooved pad
69
Figure 4.13: Microscope image of a wafer etched trenches of different depths
70
image there is a region of lower intensity where one would expect to see dark gray
peaks. It is in this region where the asperities are reaching the farthest into the wells.
There is more connectivity in the peaks in the horizontal direction than the vertical
direction. This means that the 12.5 µm deep trenches run horizontally across the
images, while the smaller 6.5 µm trenches run vertically across the images.
While the trench overlap regions are relatively easy to identify, the individual
trenches are harder to see due to the shape of the polishing pad. In an attempt to
resolve the trenches, 100 cross section in both the horizontal and vertical directions
were averaged together. The resulting average cross sections can be seen in figures
4.15 and 4.16. The average peak to valley height in figure 4.15 is 0.124±0.044 ratio
units, whereas the average peak to valley height in figure 4.16 is 0.061±0.19 ratio
units. In addition to resolving the trenches, this cross section averaging has revealed
that there is a fair amount of pad waviness across this image. There is a high point
in the pad near where the ratio is smallest: around 650 µm in figure 4.15 and around
1050 µm in figure 4.16. The corresponds to the highest concentration of green in the
top right of figure 4.14.
Wafers with multiple well depths can potentially be used for in-situ calibration.
Once the average peak to valley ratio in both directions has been determined, it can
be compared to the known average height for those cross sections. Keep in mind
that there were actually 2 types of cross sections that were averaged together for
each graph in figures 4.15 and 4.16. A cartoon of these possible cross sections can
be seen in figure 4.17. According to the dimensions in figure 4.13, 36% of the cross
sections should be inside a trench and 64% of the cross sections are not inside a
71
Figure 4.14: 3D DELIF image of a wafer with multiple step heights over a flatFruedenberg FX9 polishing pad
72
Figure 4.15: Average cross section over deep trenches
73
Figure 4.16: Average Cross section over shallow trenches
74
trench. Therefore, the actual average peak to valley height for figure 4.17 should be: