Dual Emission Laser-Induced Fluorescence (DELIF) in ChemicalMechanical Planarization

1Apone, D.,1Gray, C.,1Rogers, C.,1Manno, V.,2Philipossian, A.,3Anjur, S.,4Moinpour, M.,5Barns, C.

1Tufts University Medford, MA2University of Arizona, Tucson, AZ3Cabot Microelectronics, Aurora, IL4Intel Corporation, Santa Clara, CA5Intel Corporation, Hillsboro, OR

1 Abstract

In this paper, we present a review of the measurements that Dual Emission Laser Induced Fluorescence (DELIF) has

allowed us to make over the last 8 years. We have been able to measure many features of the slurry film between the

wafer and the pad during polishing while simultaneously measuring the frictional drag. In particular, we measured

slurry mean residence time, slurry temperature, and the slurry film thickness. All experiments are performed on

a 1:2 scaled tabletop rotary polisher with variable pad speed, wafer downforce control, and in-situ conditioning.

Mean residence times are strongly affected by conditioning. Slurry film thickness (about 20µm on average, and

accurate to within 1µm) was a strong function of macro wafer curvature as well as downforce. Slurry temperature

(accurate to 1C) increase with downforce and correlate closely to frictional drag. The resulting data are discussed for

wafers polished with a 3.1 wt percent abrasive concentration slurry solution on flat Freudenberg FX-9 polishing pads.

Correlations for friction and temperature are discussed, as well as thickness versus pressure and rotation rates.

1

2 Introduction

Chemical Mechanical Planarization is a polishing process widely used in the production of integrated circuits. Tech-

nological advances have allowed the feature sizes on these circuits to decrease greatly in accordance with Moore’s

Law, which states that the number of transistors on a chip will double every 2 years. As these circuits continue to

shrink, the challenge is to obtain a smoother, flatter surface on which to deposit the metal traces that comprise the

chip. Chips are manufactured in a layered process, so each layer must be polished to a high degree before the next set

of metal traces can be added onto the silicon substrate. The work done at Tufts University has been directed at gaining

a better understanding of the nature of this polishing process by measuring slurry properties during an actual polishing

process. In particular, in this article we will present an overview of measurements of slurry thickness, temperature,

and mean slurry residence time.

Much work has been done to try to determine the exact nature of the wafer/pad contact. There are several lubri-

cation regimes that the CMP process is thought to encompass. Modelling work indicates that it is a delicate balance

between hydrodynamic lubrication, mixed solid-liquid contact and even direct solid-solid contact. [1] The hydrody-

namic lubrication occurs when the wafer is fully supported by a layer of a slurry, with no contact between the wafer

and polishing pad. Mixed solid-liquid contact is the case when the wafer is partly supported by the pressure in the

liquid layer, and partly resting on the polyurethane asperities that rise up out of the pad. Solid-solid contact would

occur most often in either static or super high downforce experimental cases, where the wafer is entirely supported by

the pad.

More complex models have begun to take into account the pad porosity and compressibility. Modelling efforts

have been employed to predict the film thickness and removal rate based on the contact regime. Higher removal rates

can be obtained with more solid-solid contact, but with a higher incidence of wafer scratching. Changing experimental

parameters to increase the film thickness will result in less scratching of the wafer, but also a lower removal rate. [2]

Other studies have shown that the pad properties change as it soaks up water. Compression tests on the pads have

shown that the elastic modulus decreases as the pad soaks longer. The soaking also increases the pad suction pressure,

which will contribute to the contact stress on the wafer. [3]

The friction in the polishing process generates heat to dissipate energy. This increase in slurry and pad temperature

2

is important for several reasons. First, the pad properties change with temperature. The pad is a porous polyurethane

material, and becomes softer with an increase in temperature. This leads to a higher percentage of the pad contacting

the wafer, which will affect removal rates and scratching. Second, the increase in temperature will speed the chemical

reactions involving the wafer and the slurry. Generally, for an increase of 10C a chemical reaction’s rate will double.

Third, the increase in temperature (above 50C) will cause the silica particles in the slurry to clump together and act as

much larger particles. The polish removal rate, however, is dominated by the increase in chemical reactions, rather than

the mechanical factors. [4] To measure the increase in temperature experimentally, an IR camera has been employed

by many research groups. This non-invasive approach allows in situ measurements of the temperature downstream of

the polishing wafer. [4] [5]

Measuring of film thicknesses is a difficult task, since their scale (50 microns) makes ordinary measurement

techniques useless. Light Induced Fluorescence (LIF) has been used to measure thicknesses, with a camera capturing

the fluorescent image. The brightness in the image is directly related to the thickness of the film. The more fluid

there is, the more dye is present, therefore a brighter image is captured. The problem is that the initial excitation

intensity may not be even across the image. This would result in inaccurate measurements. The solution is to use a

two-dye system. Initially one dye is excited, and the light emitted from it excites the second dye. Using two cameras

to capture the image (each camera fitted with a filter for one of the dyes) and then taking a ratio of the two camera

images allows for true fluid film measurements. [6] Switching to a temperature sensitive set of dyes allows for the

optical measurement of fluid temperature. [7]

3 Experimental Setup

A Struers RotoPol-31 table top polisher unit serves as a scaled version of an industrial rotary polisher such as the

SpeedFam-IPEC 472. The polisher was fitted with a Mitsubishi Freqrol frequency modulator that allowed a working

range from 20 rpm to 300 rpm±1 rpm. The polisher has a 12 inch rotating platen that carries the polishing pad (half

of the diameter of the 24 inch commercially employed pads). To ensure the scaled setup closely replicates industrial

planarization efforts, the wafer size is scaled to match the table top polisher. The ratio of the polishing pad area,Ap,

3

to the wafer area,Aw, is held constant, resulting in a 3 inch wafer simulating a 300 mm industrial wafer.

The applied wafer pressures match the IPEC 472 and are higher than current industrial levels (from 4 - 8 psi, 26

- 52 kPa) and the relative velocities of the wafer to pad is scaled as well. The rotational velocities of the wafer,Ωw,

and pad,Ωp, are matched so that the velocity under the wafer is comparable to commercial polishing. Most runs are at

polishing pad speeds of 60 rpm, the corresponding relative velocity across the wafer is approximately 0.50 m/s. This is

typical of older polishers, with more recent polishes operating closer to 1 m/s. Measured removal rates and operating

temperatures are typical of commercial CMP as well.

For experimental procedures, an array of control and measurement systems are integrated into table top polisher.

Independently controlled parameters include wafer applied pressure, wafer rotational velocity, slurry flow rate, and pad

conditioner oscillation and rotational speeds. Throughout the experimental runs, measurement systems characterize

friction, fluid film thickness, and temperature at the wafer-slurry-polishing pad interface. Figure 1 is a schematic of

the integrated system.

[Figure 1 about here.]

3.1 Polisher Head

A Sears 20 inch drill press replaces the standard wafer carrier head mechanism of the RotoPol-31. A Dayton12 HP DC

motor controller allows drill press rotational velocities to range from 0 rpm to 160 rpm. Manual velocity adjustments

are made by adjusting the motor’s amplifier. Absolute rotational velocity is calibrated with a tachometer.

A weighted traverse is integrated into the polisher head system to achieve variable applied wafer pressures. Located

atop the drill press, this system consists of a lead screw driven carriage riding on two rails. A standard 25 lb weight

mounts on top of the carriage, positioned by a Servo Systems precision stepper motor. On one end, the traverse rests on

the rotating quill that directly attaches to the wafer while the other end is free to pivot about the main drill press support

column. Thus, the load is applied directly to the wafer with minimal external moments. Carriage position, which is

controlled by a LabVIEW interface, is calibrated before a run to wafer applied pressures using a force transducer.

4

3.2 Wafer

The DELIF optical measurement technique requires transparent BK-7 glass wafers as a replacement for traditional

opaque silicon wafers. BK-7 glass was chosen because it is structurally similar to silicon, widely available, relatively

inexpensive, and disposable. These wafers have a diameter of 75 mm (3 inch) and thickness of 12.5 mm (1/2 inch),

scaled down from 300 mm wafers. A gimbal joint mounts the wafer directly to the drill press drive shaft. In general we

found that the glass wafers are not perfectly flat, but were either slightly convex, possessing a shape that conforms to

the pad surface, (2(a)) or slightly concave, possessing a non-conforming shape as shown in figure 2(b). A profilometer

is used to perform two perpendicular line scans across the wafer diameter to characterize each wafer. If they were

more than 5µm out of flat they were discarded.

[Figure 2 about here.]

3.3 Polishing Pads and Conditioning

This research exclusively uses Freudenberg’s flat FX-9 polishing pads. The FX-9 is similar to the industry standard

Rodel IC-1000 in that it is composed of a closed cell polyurethane pore structure. In addition, the FX-9 polishing pads

are dyed black for these experiments, which minimizes the pad florescence. To date, no measurable difference due to

the dying has been found; dyed and undyed pads produce the same friction data, wear at the same rate, and polish at

the same rate. [8]

During CMP, not only do polishing pad asperities compress, but also slurry debris fills and clogs pores. The result is

a ”glazed” surface, which adversely affects planarization performance. To prevent glazing, the polishing pad surface is

conditioned by a 100 grit diamond disc to remove slurry particles while simultaneously reopening pores. The diamond

disk is mounted on a rotating gimbal type PPS (polyphenylene sulfide) carrier, and rotates as it is swept radially across

the polishing pad surface (see figure 3). Sweep and rotational rates are also controlled through a software interface.

Conditioning may be either conducted ex situ, between wafer polishes, or in situ, during the polishing process. In this

work, all conditioning is done in situ. The oscillation due to the sweeping motion of the conditioning arm periodically

starves or overloads the wafer with slurry. These periodic fluctuations in the fluid lubrication can affect planarization

5

and are clearly seen in the friction measurements.

[Figure 3 about here.]

3.4 Slurry Delivery

The slurry used is Cabot Microelectronic’s Cab-O-Sperse SC-1 in a 3.1 wt% solution with small concentrations of

dyes added to obtain fluorescence measurements. Concentrations of 1.0 g/l and 0.25 g/l of Calcein and Coumarin

4, respectively, are used for thickness measurements. 0.25 g/l of 2,3-Dicyanohydroquinone and 0.5 g/l of Brilliant

Sulfaflavin are used for temperature measurements. For the Mean Residence Time measurements, slurry is tagged

with DHPN and considered ”new” slurry while slurry tagged with Coumarin is considered ”old” slurry. The slurry

consists of 100nm fumed silica particles suspended in potassium hydroxide (KOH). In these experiments, a .3 wt%

solution was used to minimize the polish. These runs behaved similarly to the 3.1 wt% cases and did not have the

wafer changing shape over the course of the run. The 3.1 wt% solution has a viscosity of 1.13 cp1 and a shear rate

of 8200 1sec . The slurry is delivered to the CMP system by a Masterflex peristaltic pump. Its non-intrusive nature is

critical for preventing contamination, corrosion, and clogging. With 14 gauge Norprene Masterflex tubing, the pump

is capable of flow rates of up to 250 mL/min. The delivery tubing and all other slurry solution containers are opaque

to prevent photobleaching of the added dyes, a condition where a dye loses its sensitivity to light after being exposed

for a period of time. Although the actual flow rate is manually dialed in, an optical encoder on the pump enables real

time flow rate monitoring in the LabVIEW interface. The flow rates are precise to within±0.5 ml/min.

3.5 System Integration

A custom National Instruments LabVIEW program functions as the interface and controller for the scale laboratory

CMP system. Each variable may be individually controlled via the software instrument panel. All operations are

performed in the following sequence in order to avoid hysteresis and ensure repeatability.

1. Consumables preparation (pad application, wafer scans, slurry mixing)

1The viscosity values are measured at a constant temperature of 24C courtesy of Dan Airey at Cambridge Applied Systems.

6

2. Optical system setup (focus, alignment, scaling)

3. Offset microscope slide structure calibration

4. Conditioner preparation (assembly, mounting)

5. Polishing pad break-in (preconditioning of 30 minutes with a dummy wafer and undyed slurry)

6. Transition to dyed slurry

7. Introduction of experimental wafer into system

8. Experimentation with process condition variations

9. Experiment end and clean-up

10. Data post-processing

4 Measurement Techniques

In these experiments, we measure instantaneous friction force between the wafer and pad at the same time as a number

of different optical measurements. The optical measurements include slurry mean residence time, slurry temperature,

and slurry layer thickness. In this paper we will only show the highlights of each technique. For comprehensive

reviews, see work done by Coppeta [8], Coppeta and Rogers [9], Lu [10], and Cornely [11]. Friction measurements had

an estimated 3% uncertainty and the temperature measurements were accurate to 1C. The thickness measurements

were repeatable to sub-micron accuracy but the absolute measurement was accurate to within 5µm due to calibration

limitations.

4.1 Friction

The frictional force between the wafer and pad is measured to investigate the mechanical abrasion at work during

CMP. A slider force table sits under the table top polisher and on top of a 136 kg solid steel isolation table. The table

is kept from sliding in the tangential direction with a force transducer. For calibration the top plate of the slider force

table is pulled in the tangential direction by an array of hanging weights to calibrate the transducer (figure 4).

[Figure 4 about here.]

7

Repeated friction force measurementsFf are typically the average of one thousand points sampled at 100 Hz

during steady state planarization conditions. This was fast enough to sample all of the modes of the force and long

enough to capture about 10 platen rotations. Results are represented here as the raw friction force per unit wafer area

(kPa). The friction data is normalized by the applied wafer pressure to obtain the dimensionless coefficient of friction

Cf .

Cf =Ff

Fdf(1)

In equation 1,Cf is the coefficient of friction,Ff is the force of friction, andFdf is the applied wafer pressure at wafer

center. The friction measurements have fluctuations due to the effects of conditioner oscillation, system misalignment,

and pad characteristics [10].

4.2 Dual Emission Laser Induced Fluorescence Techniques

The bulk of the measurements reported in this work are using Dual Emission Laser Induced Fluorescence (DELIF).

DELIF is a relatively non-intrusive optical technique for measurement of a passive scalar. Fluorescence is the emis-

sion of electromagnetic radiation by a substance whose molecules are excited into higher energy states through the

absorption of a certain wavelength of radiation. In DELIF, light in the visible spectrum is emitted by the molecules of

dyes excited by ultraviolet light. The fluorescence intensity emitted by a dye is the function of excitation intensity and

amount of absorbed light. Quantitatively, it may be estimated using

If = Φ · Ie ·Abs, (2)

whereIf is the measured fluorescence intensity,Φ is the quantum efficiency of the dye in solution,Ie is the intensity

of the excitation energy source, andAbs is the amount of light absorbed by the dye. In one dimension, the fluorescence

intensity of a point of dyed solution excited by a single beam of light may be represented by

If (b) = Ie ·Af · Φ · ε(λ) · C · dl, (3)

8

whereb is the distance travelled by the laser beam through the dye solution,If is the measured fluorescence intensity

at the distanceb through the solution,Ie is the intensity of the excitation energy source at the same distanceb through

the solution,Af is the fraction of the fluoresced light collected,ε is the molar absorptivity as a function of wavelength

λ or pH in the case of pH dependent dyes,C is the molar concentration of the dye, anddl is the length of the sampling

volume along the path of excitation. [12]

Single dye techniques for absolute measurements tend to be less reliable because they are strongly dependent on

excitation light intensities. Shadows, light variations in time, and non-uniform light sources all influence the resulting

florescence intensity. DELIF is a much simpler alternative to developing complex calibrations to limit some of the

factors. In DELIF, two dyes are dissolved in solution. Normalization of one dye to the other eliminates many of

the issues that would otherwise require calibration. Fluorescence intensities of a passive scalar dependent dye are

normalized by a non-passive scalar dependent dye. With respect to CMP, passive scalars include fluid film thickness

and temperature. Equation 3 becomes,

Fluorescence Ratio =I1f

I2f=

ε1(λ) · C1 · Φ1

ε2(λ) · C2 · Φ2. (4)

As a result of normalization, the effect of excitation energy source variations, light reflections, and other non-uniformities

along the path of lightdl on fluorescence measurements are minimized. Figure 5 depicts the four principle steps in the

DELIF measurement technique.

[Figure 5 about here.]

Although equations 3 and 4 physically quantify dye fluorescence, they do not account for other dye fluorophor

behavior such as photoquenching or photodegradation. Photoquenching occurs when dye fluorescence responds non-

linearly to linear changes of excitation energy, a result of too much light. Photodegradation is decomposition of

dye fluorophor by the excitation energy and is due to prolonged exposure to light. All dyes have some degree of

photodegradation and can be typically characterized by a time dependence due to decomposition.

In addition to photoquenching and photodegradation, many other factors complicate qualitative measurements

with dye fluorescence. Chemical changes such as pH, temperature, or excitation energy variations directly impact

9

dye fluorescence spectra behavior. Several complications also arise with excitation energy fluctuations, physical light

reflections, and environmental noise such as ambient light. Coppeta and Rogers [9] divide these effects into three

different types of errors. Type I conflicts arise when dye A and dye B have emission spectra that overlap, so discerning

the source of the fluorescence intensities is impossible. Type II conflicts arise when the fluorescence spectra of dye

A overlaps with absorption spectra of dye B while the absorption spectra of dye B remains unaffected by changes

in the passive scalar being measured. As a result, dye B absorbs the emission of dye A with a dependent being the

path length travelled by the emission of dye A through dye B. The longer the distance, the greater the opportunity

for absorption by dye B. Due to this attenuation of dye A and none so for dye B, their ratio is path length dependent.

Finally, type III conflicts are similar to type II except for a second variable: the absorption spectra of dye B also varies

with a passive scalar, such as temperature. All dyes used in these measurements are were chosen to minimize these

three errors.

4.3 Image Acquisition

The fluorescence of the two dyes is individually captured by two independently filtered digital cameras. A custom

optics system is used to ensure that the cameras image the same exact spatial fluorescence. Internal components such

as the beam splitter and filters are shown in figure 6(b).

[Figure 6 about here.]

Each image must also be aligned both orthogonally and rotationally to ensure ratioing of the same fluorescence area

in post-processing after capture. Prior to the experiment, orthogonal alignment is achieved through hardware and

software manipulation. After the Region of Interest (ROI) is selected, both identical images, are cropped and shifted

horizontally and vertically until the are perfectly on top of each other. Rotational alignment is achieved through manual

adjustment of a rotation stage.

The pair of 768 x 512 pixel CCD Photometrics Sensys cameras are capable of a linear response to light intensi-

ties down to less than 0.1% of their entire 12-bit dynamic range. Lengthy exposure times of typically 3000 ms per

fluorescence image were necessary due to the relatively low excitation energy of the ultraviolet lamps. ”Binning” or

10

combining adjacent pixels to increase overall image imaging intensities may be also needed despite a compromising

of resolution. Due to the long exposure time, each image is a temporal average of the entire length of exposure. Each

image is then spatially averaged to obtain a representative mean ratio. Typically, a steady state measurement is the

calibrated mean of five such averaged image ratios, or a total of 15s of exposure.

4.4 Mean Residence Time

Fresh slurry must constantly be replacing the used, photobleached slurry in order to yield consistent results in our

temperature and thickness experiments. If slurry kept being reused, it would eventually lose its fluorescent ability

and distort the recorded thicknesses. Tests were run to determine if there was a difference between in situ and ex

situ conditioning. Fresh slurry is tagged with DHPN. Slurry already on the platen was tagged with Coumarin and

considered ”old” slurry. Images were then taken every 2 seconds, and the ratio of ”new” to ”old” slurry was compared

and analyzed. [8]

4.5 Slurry Film Thickness

The CMP system approaches boundary lubrication as the fluid film layer between wafer and polishing pad thins,

whereas hydrodynamic lubrication is approached as the layer thickens. DELIF is utilized to characterize the fluid film

thickness. A general overview is presented here, but a complete description of these measurements, comprehensive

application, and limitations are described by Coppeta [8], and Lu [10]. The approach here is to take advantage of a

type 2 error to measure the total amount of dye at a pixel in an image.

Two main requirements for thickness measurement need to be met:

• Fluorescence of dye A (emitter) overlaps with absorption of Dye B (absorber)

• Dye B absorption spectra unaffected by changes in passive scalar

Calcein (Fluorexon) was chosen for the absorber, and Coumarin 4 (4 Methyl-Umbelliferone) for the emitter. Con-

centrations of 1.0 g/l and 0.25 g/l of Calcein and Coumarin 4, respectively, are used. These were chosen to maximize

the fluoresced light in the camera images. Figure 6 shows the absorption and emission spectra of both dyes along

11

with chosen filter bands. The calcein fluorescence was measured divergent from Calcein’s peak emission to minimize

capture of Coumarin’s slightly overlapping fluorescence tail. Therefore, the DELIF ratios the amount of unabsorbed

Coumarin emission to the amount absorbed (as measured by the Calcein emission). This is proportional to the con-

centration of the Coumarin and Calcein and hence yields a direct measure of the amount of slurry in the measuring

volume.

[Figure 7 about here.]

With this setup, film thickness increases at the wafer-slurry-polishing pad interface lead to increases of emission in

both dyes. The net Coumarin fluorescence,I2f , will decrease because of absorbtion while Calcien fluorescence,I1f ,

increases before being viewed by the cameras. The measured fluorescence ratio, as described by equation 4, increases

in accordance with the film thickness. An absolute film thickness measurement accurate to within 5µm with a relative

accuracy of about 1µm is then attained by calibration. These limitations are not imposed by the technique, but rather

by the calibration rig and could be improved if needed.

Although the technique accurately measures the physical fluid film thicknesses, the rough surface along with

entrapped dyed slurry of the polishing pad make an absolute measurement difficult (within 5µm). As a result, there

is no true absolute polishing pad ”surface,” so all thickness measurements are from the wafer surface to some mean

depth within the polishing pad, since the technique really measures the volume of fluid at a single pixel location on

the camera.

As with typical fluorescence based measurement techniques, outside parameters, such as pH and temperature, in-

fluence fluorescence. Ideally, any variations in the flow field of measurement would be normalized by the two dye

system. The dyes also have non-linear fluorescence intensity, especially at extreme thicknesses. Photodegradation of

Calcien leads to a mean fluorescence ratio drop of almost 40% within 5 minutes of ultraviolet lamp excitation. How-

ever, there is a constant inflow of new dyed slurry solution during CMP with complete replacement slurry underneath

the wafer in 70 seconds [8]. In addition to dye photodegradation, polishing pads can fluoresce similarly as well,

altering its inherent fluorescent behavior. This adversely affects measurement continuity by changing the background

noise taken into account by calibration.

12

4.6 Temperature

Temperature measurements at the wafer-slurry-polishing pad interface take advantage of DELIF and corresponding

ratio normalization. The interfacial temperature importance lies in its direct influence upon the chemical reactions

between wafer surface and slurry. Material removal depends on the chemical reactions, which are directly affected

by temperature. The dyes chosen must be temperature rather than thickness dependent. A brief overview of the

technique will be presented here; a fully detailed technique development and implementation description can be found

in Cornely’s work [11].

The two temperature sensitive dyes, 2,3-Dicyanohydroquinone (DHPN) at a concentration of 0.25 g/l and Brilliant

Sulfaflavin at a concentration of 0.5 g/l, served as the emitters in slurry solution. Brilliant Sulfaflavin increases in

fluorescence intensity as temperature rises, while DHPN behaves in the opposite manner. In regard to equation 4,

an increase in the interfacial wafer-slurry-polishing pad temperature results in an overall decrease in theI1f (DHPN)

to I2f (Brilliant Sulfaflavin) fluorescence ratio. Due to the near identical nature of the temperature and fluid film

thickness technique, the twin camera optical system remained exactly the same all the way down to the filters. DHPN

was filtered at the 455 nm to 500 nm bandwidth while Brilliant Sulfaflavin was filtered to the 550 nm to 600 nm

bandwidth.

Two calibrations are necessary for the temperature measurements. First, the fluid film thickness dependence of the

dyes must be eliminated. Second, the measured dye fluorescence ratios must be calibrated to absolute temperatures.

For simultaneous fluorescence and absolute temperature measurements, a thermocouple is applied in parallel with

DELIF. By plotting the absolute temperatures with the ratio of the measured dye fluorescences, a linear relationship

between the two is achieved. A range of temperatures from 20C to 40C is achieved through gradual heating of the

slurry inflow. The fluorescence ratios are plotted against the absolute measurements to obtain a linear fit. Calibration

errors were found to be on the order of 0.33C, with a further 0.5C error resulting from the thickness correction.

Each fluorescent image is the spatial average of a 3000 ms temporal average (camera exposure time). Finally, each

data point reported is the mean of five of those temporally and spatially averaged images implying 15 second averages

of 393,216 pixels. The standard deviation of temperature fluorescence ratios in a temporally averaged image is less

than±0.25C, and overall measurement uncertainty is±1C. [11]

13

An important deviation from the fluid film thickness technique is the transient nature of temperature in CMP.

This renders applicable characterizations of in situ temperature difficult. Data show hysteresis due to continued pad

heating. Up to 5 minutes are required to reach steady state temperatures, but an industry oxide polish typically lasts

only 2 minutes. Therefore each temperature measurement was taken after 1 minute of polishing to approximate half

the typical oxide polish time.

5 Results and Discussion

5.1 Mean Residence Time

The first measurements made were looking at how long a tagged bit of slurry was exposed to the wafer during polish.

The mean residence time measurements show that after 70 seconds we have totally fresh slurry under the wafer (flow

rate of 35 cc/min). Coppeta examined in situ versus ex situ conditioning and found that the in situ conditioning slightly

sped the replacement rate. The ex situ conditioning runs took about 5 seconds longer to totally replace the slurry under

the wafer. This information is important for removing polish waste. This is also important for DELIF measurements

because the slurry will eventually photobleach, reducing its effectiveness. Fresh slurry must replace bleached slurry

under the wafer fast enough so that our fluorescence intensity measurements remain accurate. If an image were taken

while ”old” slurry was underneath the wafer, the thickness measurement could be thrown off, since the dyes will not

necessarily photodegrade at the same rate.

[Figure 8 about here.]

5.2 Film Thickness

After measuring how quickly the fluid is replaced, our next step was to measure the thickness of that slurry layer. We

found that layer to be strongly dependent on the wafer curvature. The thickness data (figure 6) show that the thickness

decreases with downforce and increases with pad speed for a convex wafer. The higher downforce should ”crush” the

liquid layer and the pad asperities, causing the DELIF setup to measure a thinner layer. As the pad rotational velocity

increases, more polishing slurry gets dragged under the wafer and the result is a thicker layer of fluid. The fluid film

14

responds linearly to increases in downforce and pad speed, though the downforce-thickness relationship is an inverse

one.

The problem gets a bit more complicated for the concave wafer since the main contact area is the circumference.

Instead of riding up on a ”cushion” of slurry like a water skier, this contact region ”pushes” through the slurry like a

snow plow. This keeps the overall fluid film thickness fairly constant with changes in pressure and velocity.

[Figure 9 about here.]

5.3 Friction

One can correlate the thickness changes with changes in the frictional measurements. For the convex wafer, the

COF increases slightly with wafer downforce, and decreases with pad speed. As the fluid gap gets thicker less of

the pad material is in contact with the wafer, which shows in lower frictional measurements. Conversely, as the gap

gets thinner the pad’s asperities increase the measured friction. As the wafer downforce increases, more of the pad’s

asperities come in contact with the wafer, increasing the COF. If one were to speed up the pad, the wafer would ride

up on a thicker layer of slurry, decreasing the COF. As was the case with thickness measurements, concave wafer

frictional measurements change little with changes in pad velocities when compared to the convex wafers. Lu [10]

mapped these measurements on a Streibeck curve and also looked at how these values changed as the wafer polished

from concave to convex.

[Figure 10 about here.]

More interesting that the mean frictional data, however, is the spectral content of the friction measurements. The

convex wafer shows much lower frictional forces, mainly due to the fact that it rides on the cushion of slurry much

more easily than the concave wafer. The concave wafer’s high frictional values can be explained by the leading edge

digging into the pad much more violently. The stick slip phenomena become less apparent as the film layer gets

thicker. Smoothly riding on a thicker layer of fluid would keep the leading edge from coming in contact with the pad

as much. This is much easier for the convex wafers to do, explaining the much cleaner frequency spectrum.

[Figure 11 about here.]

15

5.4 Temperature

Friction is also strongly correlated to the local slurry temperature. We measure temperature both in front of the

wafer (with a thermocouple) and under the wafer (DELIF). Both measurements were strongly correlated and we will

therefore only report the under-wafer measurements. [11] The temperature versus pressure chart begins to show the

relationship between friction and temperature. The concave wafer shows higher temperatures, both of the bow wave

and under the wafer, than the convex one due to its higher friction values. As the downforce increases, both wafers

increase their respective temperatures. Energy is dissipated by friction as heat, so as experimental runs are carried out

that have higher friction values, those runs will have higher temperatures.

The temperature versus rotation rate graph shows similar data to the temperature versus pressure graph. The

concave wafer shows higher temperatures here too, as it’s digging into the pad harder. As the speed increases the

temperatures go up as well. This is not due to an increase in COF, but rather the increase in the number of microscopic

collisions between wafer, pad, and silica particles. Doubling the rotation rate will cause twice as many pad asperities

and silica particles to flow past the wafer. This increase in energy is then dissipated as heat.

[Figure 12 about here.]

The final graph shows a linear correlation between friction force and the temperature for both wafers. As the

frictional force increases, so does the temperature, which is to be expected as the friction is what is responsible for the

increase in temperature. The concave wafer has higher temperature and friction data than the convex wafer does, which

is in keeping with previous data. The concave wafers have higher temperatures both in the bow wave and under the

wafer, as well as higher friction measurements than the convex ones. This implies that average friction measurements

can be used for estimating slurry temperature under the wafer, regardless of wafer shape.

[Figure 13 about here.]

16

6 Conclusion

A DELIF technique for gathering optical measurements during a CMP polish was presented. Fluid film thicknesses

were found to be between 20 and 30 microns depending on specific run parameters. Increasing the applied pressure

would decrease the film thickness, while increasing the velocity would increase the film thickness. Mean COF mea-

surements were found to range from .05 to .2 depending on wafer shape and run parameters. The concave wafers

show higher friction values than the convex ones, due to the more violent stick slip action, as shown in the frequency

spectrum. Slurry under the wafer is totally replaced every 70 seconds. The temperature under the wafer was shown

to range from 27 to 37C. The bow wave temperature and friction measurements are both excellent indicators of the

temperature underneath the wafer. Temperature rises slightly with rotation rate (more for concave than convex) and

increases with downforce. Finally, an increase in the friction forces was shown to increase the temperature under the

wafer linearly.

References

[1] S R Runnels and L M Eyman. Tribology analysis of chemical-mechanical polishing.Journal of the Electro-

chemical Society, 141(6):1698–1701, 1994.

[2] D G Thakurta, C L Borst, D W Schwendeman, R J Gutmann, and W N Gill. Pad porosity, compressibility

and slurry delivery effects in chemical mechanical planarization: Modeling and experiments.Thin Solid Films,

336:181–190, 2000.

[3] S H Ng, R Hight, C Zhou, I Yoon, and S Danyluk. Pad soaking effect on interfacial fluid pressure measurements

during cmp.Transactions of the ASME, 125:582–586, 2003.

[4] H J Kim, 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, 130-131:334–338, 2002.

[5] H Hocheng and Y L Huang. In situ endpoint detection by pad temperature in chemical mechanical polishing of

copper overlay.IEEE Transactions on Semiconductor Manufacturing, 17:180–187, 2004.

17

[6] C H Hidrovo and D P Hart. Emission reabsorption laser induced fluorescence (erlif) film thickness measurement.

Measurement Science and Technology, 12:467–477, 2001.

[7] J Sakakibara and R J Adrian. Whole field measurement of temperature in water using two-color laser induced

fluorescence.Experiments in Fluids, 26:7–15, 1999.

[8] J R Coppeta.Investigating Fluid Behavior Beneath a Wafer During Chemical Mechanical Polishing Process.

PhD thesis, Tufts University, 1999.

[9] J Coppeta and C Rogers. Dual emission laser induced fluorescence for direct planar scalar behavior measure-

ments.Experiments in Fluids, 25:1–15, 1998.

[10] J Lu, C Rogers, V Manno, A Philipossian, S Anjur, and M Moinpour. Measurements of slurry film thickness and

wafer drag during cmp.Journal of the Electrochemical Society, 151, 2004.

[11] J Cornely, C Rogers, V Manno, and A Philipossian. In situ temperature measurement during oxide-chemical

mechanical planarization.Proceedings for the MRS Spring Meeting, Santa Clara, CA, 2003.

[12] G G Guilbault.Practical Fluorescence: Theory and Practice. Marcel Dekker, 1976.

18

List of Figures

1 Laboratory scale tabletop CMP polishing system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2 Definition of wafer shapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Pad conditioning system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Schematic of the friction force slider table calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5 Schematic of the Dual Emission Laser Induced Fluorescence (DELIF) technique . . . . . . . . . . . 24

6 Dual camera system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

7 Dye Emission and Absorption Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

8 Mean Residence Time of Slurry under Wafer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

9 Thickness Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

10 Friction Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

11 Stick Slip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

12 Temperature versus Rotation Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

13 Correlation of Friction Force and Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

19

Figure 1: Laboratory scale tabletop CMP polishing system

20

(a) Convex wafer

(b) Concave wafer

Figure 2: Definition of wafer shapes

21

(a) Front view (b) Side view

Figure 3: Pad conditioning system

22

Figure 4: Schematic of the friction force slider table calibration

23

Figure 5: Schematic of the Dual Emission Laser Induced Fluorescence (DELIF) technique

24

(a) Nikon enlarging lens with an optics package mounted to 2 Photomet-rics Sensys cameras

(b) Filter stacks and a cubic 50/50 beam splitter housed inside the opticspackage

Figure 6: Dual camera system

25

Figure 7: Dye Emission and Absorption Spectra

26

Figure 8: Mean Residence Time of Slurry under Wafer

27

Figure 9: Thickness Data

28

data2.eps

Figure 10: Friction Data

29

slip2.eps

Figure 11: Stick Slip

30

Figure 12: Temperature versus Rotation Rates

31

vs fric.eps

Figure 13: Correlation of Friction Force and Temperature

32