in-situ measurement of pressure and friction during cmp of ... · measured friction coefficient...

Submitted to the Journal of the Electrochemical Society

In-Situ Measurement of Pressure and Friction During CMP of Contoured Wafers

A.M. Scarfob, V.P. Mannoa, C.B. Rogersb, S. Anjurc and M. Moinpourc

a – Corresponding Author, Department of Mechanical Engineering, Tufts University, Medford, MA 02155,

617-627-2548, FAX-627-3819, [email protected] b - Corresponding Author, Department of Mechanical Engineering, Tufts University, Medford, MA c – Cabot Microelectronics, Aurora, IL d – Intel Corp., Santa Clara, CA

Abstract In this paper we document in-situ fluid film pressure and interfacial friction measurements during

Chemical Mechanical Planarization (CMP) over a range of applied loads and relative pad/wafer

velocities. The slurry film pressure beneath contoured test wafers is measured using a novel experimental

setup that enables dynamic data collection. The friction data have a repeatability of approximately 10%.

The uncertainly of the pressure measurements and the computed down forces are ±20.7 kPa (±3 psi) and

20%, respectively. The data indicate that wafer shape, specifically global curvature, is a significant factor

in determining the lubrication regime during CMP. Full hydrodynamic lubrication, in which the slurry

fluid film supports the entire applied load, was not realized for CMP of either concave (center high) or

convex (center low) wafers. The data for concave wafers show that –6% to 37% of the applied load is

supported by the slurry film, where the negative sign indicates suction conditions that were obtained at the

lowest applied load condition. CMP of convex wafers is found to operate closer to full hydrodynamic

lubrication, with the fluid layer supporting 36% to 64% of the applied downforce. In all cases, the

measured friction coefficient decreased as the support of the fluid layer increased (higher positive

pressures). CMP of concave wafers is more sensitive to changes in applied downforce, while the convex

wafer type was most affected by changes in the wafer/pad rotation speed, which in turn determines effective

slurry film velocity beneath the wafer. Overall, the CMP conditions seen in these scaled experiments

operate primarily in the partial lubrication regime shifting closer to hydrodynamic lubrication for convex

wafers at the high load, high speed conditions.

1

mailto:[email protected]

Submitted to the Journal of the Electrochemical Society 1. Introduction

Chemical Mechanical Planarization (CMP) is an integral step in the manufacture of multilevel

integrated circuits. In oxide CMP, a rotating silicon wafer substrate is pressed against a rotating polishing

pad as chemically active, abrasive slurry is continually injected at the wafer/pad interface. CMP is superior

to other planarization methods due to its ability to achieve high levels of both local and global planarity

across the wafer surface. Due to its widespread use, a thorough understanding of the relationship between

CMP process conditions and the physics of material removal mechanisms is desirable. Much of the

uncertainty in obtaining this correlation relates to characterizing the tribology dynamics and lubrication

regimes that obtain for various CMP operating conditions.

CMP lubrication regime depends largely on how the applied load is distributed between the fluid

film layer and the pad asperities. In hydrodynamic lubrication, the fluid film is continuous with negligible

pad-wafer contact. As such, the fluid pressure balances the entire applied load. Material is removed in the

hydrodynamic regime through a combination of fluid induced erosion and slurry particle abrasion. At the

other extreme, boundary lubrication, the fluid film layer is thinner and there is significant solid-solid

contact between the wafer and pad, especially the pad asperities. In this case, material removal is due

primarily to the mechanical abrasion of the asperities against the wafer surface. The abrasion is enhanced

by chemical activation of the wafer surface and silica slurry particles entrapped in the pad asperity

interstices. The consensus hypothesis is that most practical oxide CMP operates somewhere in between

these two extremes in what is known as the partial lubrication regime.

Several investigations have focused on better understanding the behavior of the fluid film layer

during CMP. Levert et al. [1] studied the role that surface properties play in defining the lubrication regime

during CMP. They found that, for an acrylic and semi-permeable pad, the film layer thickness increased

with increasing relative wafer/pad speed and applied load. However, no significant slurry film developed

for the permeable pad sample. In another study, Shan et al. [2] collected pressure data between a rotating,

soft polyurethane pad and a stationary opposed steel disk to determine how the applied force is distributed

at the interface. They found that the leading (i.e. slurry injection side) two-thirds of the wafer experienced

sub-ambient pressure, while positive load-bearing pressures were only obtained in the vicinity of the wafer

trailing edge. Bullen et al. [3] used a pressure measurement system that was a prototype of that used in this

2

Submitted to the Journal of the Electrochemical Society work. They collected data during static (wafer not rotating) and dynamic (wafer rotating) CMP. Although

trends were consistent with Shan et al., Bullen’s results show higher pressures for the static cases. As such,

they conclude that the static method is not a good approximation of the actual CMP process.

Our group has focused on exploiting a combination of optical and ancillary experimental

techniques to measure directly the slurry film during CMP. The primary optical technique used is known as

Dual Emission Laser Induced Fluorescence or DELIF [4]. Using DELIF, Lu et al. [5] measured the slurry

fluid film thickness at various applied loads, wafer/pad relative velocities, and wafer curvatures. Wafer

curvature, which is characterized qualitatively as convex (wafer edges curving away from pad surface) or

concave, is determined by the relative orientation of the wafer and pad surfaces. As the current DELIF

implementation requires transparent wafers, BK7 glass wafers are used as surrogates for silicon wafers. Lu

et al. found that, for convex wafers, the fluid layer thickens with increasing speed and decreasing

downforce, resulting in a decrease in friction. The data were opposite for concave wafers; the film actually

grew thicker in the area of inquiry as the downforce increased. Note that since the slurry region near the

outer edges of the wafer is unavailable for measurement, no comment on the overall film thickness

behavior can be made.

In this paper, we expand on the earlier work of Lu et al. In this study, fluid film pressure and

interfacial friction measurements were collected during dynamic CMP using contoured Plexiglas wafers.

As is described in the next section, Plexiglas is used rather than BK7 glass due to the wafer machining

required for the pressure data collection. Each wafer is classified based upon its shape and data are reported

over a range of applied downforces and relative pad/wafer rotation speeds. The net force on the wafer is

compared with the applied force to determine the percentage of the load supported by the fluid film. In

addition, the interfacial drag data are analyzed to elicit how certain process parameters change the degree of

contact between the pad and wafer.

2. Experimental Setup

Figure 1 is a schematic diagram of the test setup. The rig is a Struer’s Roto Pol tabletop polisher,

built on a 1:2 scale based on a SpeedFam IPEC 472 rotary polishing tool. A 30.5 cm (12 inch) diameter

removable aluminum platen serves as the pad carrier. A modified drill press, which also acts as the wafer

3

Submitted to the Journal of the Electrochemical Society carrier, is to allow variable downforce and wafer rotation speeds. A weighted traverse mounted atop the

drill press provides variable downforce ranging up to 50 kPa (7 psi). Wafer rotational speeds can be set

from 0–120 rpm ± 1 rpm. The pressure measurement hardware is also mounted concentrically on the drill

press spindle.

The polisher is positioned on a friction force detector consisting of two parallel plates separated by

a pair of linear rails that allow movement in only one direction (tangential to the polisher motion). A load

cell is mounted between the plates to detect any motion of the polishing pad relative to the wafer. Since the

drill press is bolted to the floor, the relative motion of the polisher gives a measure of the friction between

the pad and wafer. Freudenberg FX-9 polyurethane polishing pads, which are relatively hard with closed

cellular structure, are used in the data reported. In-situ pad conditioning is employed using a 5.1 cm (2

inch) diameter 163 µm diamond grit wafer, which rotates and oscillates across the pad to prevent pad

glazing and to keep pad surface conditions relatively constant.

A custom pressure measurement device was developed for this study to enable in situ dynamic

fluid film pressure measurements (see Figure 2). More details on the design, fabrication and operation of

the pressure measurement system are available in Scarfo [6]. The device consists of three basic assemblies,

each performing a specific function: pressure detection, position sensing, and data transfer. Pressure

detection is achieved using Omega PX-203 series transducers. Each transducer has a 1 ms response time

and can measure pressure ranging from 0-207 kPA or 0-30 psia providing a quantitative measure of suction

force. Incorporating all factors of the measurement uncertainty is ± 21 kPa or ± 0.3 psi. The transducers are

connected to the wafers via Tygon tubing. Special adapters enable removal of trapped air in the tubing

extending from each transducer to the wafer in order to reduce artificial damping of the pressure

fluctuations.

The position sensing system consists of a Hall switch magnetic sensor and a magnet mounted on

the side of the rotating base. The Hall switch, which produces an electrical signal due to the voltage

changes when the embedded magnet passes in its vicinity, is secured to the stationary portion of the

polisher. Since both the pad and wafer rotate during CMP, data and power leads are secured to a slip ring,

mounted concentrically on the rotating bit. The signals are passed over sliding metal contacts in the slip

ring and are wired directly to the computer. Computer-based data acquisition software is used to record the

4

Submitted to the Journal of the Electrochemical Society pressure and the position signals simultaneously. Data post-processing allows the pressure data to be

correlated to the angular position in the rotation.

Typically, BK7 glass is used in our DELIF set up. However, BK7 glass is brittle and difficult to

machine without inducing cracks. Instead, Plexiglas is employed as the wafer material. The Plexiglas

wafers are tapped in six locations positioned 60° apart, spaced linearly from 1.3 to 3.3 cm from the wafer

center (see Figure 3). These taps are needed to allow pressure communication from the wafer/pad/slurry

interface region of interest to the corotating pressure transducers. Each tap consists of a small hole

extending approximately halfway though the wafer thickness. A larger tapped hole mates to meet this

smaller hole. Barbed tubing connectors are screwed to enable tubing attachment. The current

implementation of the pressure rig only allows three simultaneous pressure measurements at one time.

Therefore, a data run is conducted in two sequential measurement periods. After half of the data are

acquired, the device is removed and reconfigured to allow sampling at pressure tap numbers 4, 5, and 6

after the data acquisition from tap numbers 1, 2 and 3 is completed.

As wafer shape affects the lubrication properties during CMP, each of the 25 test wafers were

scanned using a VEECO Dektak V 200-Si stylus profilometer. Two perpendicular line scans crossing the

wafer center are used to classify the wafer shape. Since the wafers were manufactured from bulk Plexiglas

1.27 cm (0.5 inch) sheet stock, surface irregularities exist in the machined surfaces. Wafers are classified

into one of three categories: convex, indeterminate or flat, or concave. As depicted in Figure 4, a convex

wafer is bowed outward toward the pad while the opposite is the case for concave wafers. In this

investigation, 5 wafers are classified as concave or type A, with a center to edge bow ranging from 5-15

µm. 10 wafers fall into the indeterminate category (type B), neither clearly convex nor concave or close to

flat, and 10 wafers are designated as convex or type C, with 5-15 µm center to edge convexity. In practice,

CMP slurries include suspensions of small silica particles that enhance material removal. Also, slurry

particles clog the small holes in the tapped wafers. As this investigation required non-varying wafer

curvature, commercial slurries are not useful since they would slowly polish the Plexiglas and change the

curvature. Instead, NaOH buffered water (pH~11) is used as the polishing “slurry”.

5

Submitted to the Journal of the Electrochemical Society 3. Results and Discussion

Each wafer data set consists of a series of ring pressure measurements taken at the six radial

positions depicted in Figure 3. Data for a total of 25 rotations were collected for each wafer tested at each

of the six operating conditions. The data acquisition system was programmed to sample 1800 points per

rotation with every five sequential data samples averaged to yield a new of 360 data. The data for each of

the three wafer curvature categories are then averaged to yield the results presented herein. Pressure

measurements are normalized by the applied downforce and are plotted versus the normalized wafer radial

position. Friction measurements are also reported for all wafers at each operating condition. These data are

also segregated and averaged based on wafer curvature classification.

Figure 5 displays pressure data for each wafer type at an applied down load pressure of 27.6 kPa

(4 psi). Results differ significantly among the three wafer shapes. For the concave wafer type A,

subambient pressure is obtained near the middle portion of the wafer. Pressure seems to start out slightly

positive near the wafer edge, but a vacuum is created upon moving closer to the wafer center. The pressure

profile shifts slightly more positive as the relative speed is increased, but the effect is minimal. In

qualitative agreement with Levert et al. [1], only positive pressure is observed for convex wafer type C,

increasing parabolically toward the wafer center. Relative wafer speed changes have more of an effect in

the case of convex wafers, increasing the maximum pressure observed from 75% to 130% of the applied

load. The behavior of the indeterminate or near flat wafer type B falls between the behaviors of the other

two wafer shapes. The pressure is positive over all conditions for wafer type B and increases at the wafer

center. However, there is a slight dip in pressure upon moving inward from the outermost tap.

Analogous data for a higher applied load (41.4 kPa or 6 psi) are reported in Figure 6. The increase

in load seems to cause a transition from subambient to positive pressure for wafer type A over all speeds.

The maximum fluid pressure observed corresponded to nearly 80% of the applied load compared to a

maximum of 20% for the lower load condition. Hence, the fluid film never fully supported the applied load

under any of the conditions tested. The load not supported by the film is presumed to be supported through

wafer/pad contact. Although the slurry film pressure moved into the positive range, the effect of the

curvature is still reflected in the shape of the pressure profile. Upon moving in from the edge, the pressure

dips lower but then builds up again. The increase in load seems to result in a more classic parabolic

6

Submitted to the Journal of the Electrochemical Society pressure profile. The normalized pressure under wafer type C increased with applied load, with the

maximum local pressure at the highest speed reaching 170% of the average applied pressure, compared to

130% at the lower applied load condition.

The average normalized friction for each of the three wafer types is presented in Table 1 for all

six operating conditions. It is clear that the friction is greatest for the concave wafer type (A) and least for

the convex wafer type (C). Increasing the relative wafer/pad speed causes a reduction in the friction

coefficient for all wafer types. The increase in the applied pressure from 27.6 kPa to 41.4 kPa decreased

friction level for wafer type A, but has little influence on friction for the other two wafer shapes.

The net force exerted on each wafer was computed in order to represent the pressure data as a

single, integral parameter. The calculation procedure involved dividing the wafer surface area into a series

of six concentric rings, each corresponding to an annular region associated with each of the pressure tap

radial position. The measured fluid pressure at a particular tap location is then multiplied by its respective

area and the force subtotals are summed to yield a net opposing force exerted on the wafer. The results,

calculated as the % of applied load, are displayed in Table 2. Once again, the results differ among each

wafer type. Wafer type A seems to border on boundary lubrication at the lower applied load condition

despite increasing the speed. However, a clear transition occurs when the load is increased, causing the net

opposing fluid force to build to over 30% of the applied downforce. As the wafer shape transitions from

concave to convex (i.e. type A to type B to type C), the fluid film supports a greater percentage of the

applied load. However, fully hydrodynamic lubrication is not observed under any of the conditions tested,

with only 60-65% of the applied load borne by the fluid film at the data extrema.

Figures 7 and 8, which are complementary to the data reported in Table 2, illustrate the sensitivity

of each wafer type to changes in applied load and rotation speed. Wafer type A is most sensitive to the

changes in downforce. One possible explanation is that as the downforce is increased, the outer edge of the

concave wafer digs into the pad causing the pad to deform. The pad, in turn, conforms to the bowed area at

the wafer center, creating a converging gap. As a result, positive pressures develop. This transition is also

marked by a 26% reduction in the normalized friction force, often defined as the coefficient of friction or

COF (see Figure 8). As mentioned earlier, these observations must be assessed with the knowledge that the

outer regions of the slurry film are out of measurement range.

7


Increasing wafer rotational speed has the greatest effect on the fluid flow under wafer type C.

Even at low speeds, a parabolic pressure profile develops due to the convex geometry of the flow passage.

As the relative velocity increases, the load carrying capacity of the fluid film also increases, causing the

lubrication regime to move closer to fully hydrodynamic. As expected, the friction coefficient drops (by

43%) under these conditions. Qualitatively, it seems that an increase in speed causes the convex wafer to

hydroplane on the continuous fluid layer.

4. Conclusion

The data reported in this paper add insight into the lubrication regime of the CMP process under

various operating conditions. In particular, the interplay of downforce, relative pad/wafer velocity and

global wafer contour is reported. The in-situ, concurrent fluid film pressure and integrated friction data are

useful not only for characterizing CMP operations but also as validation data for simulation model

developers. Certain data trends are noted by combining this pressure data with the measured friction

coefficients for each wafer type. Additional data are also available in the work of Scarfo [6]. In general, it

appears that all wafers tested operated in the partial lubrication regime over most conditions, with the fluid

film supporting a fraction of the applied load. At low loads, the concave wafer type exhibited suction and

high friction, most closely representative of the full boundary lubrication regime. However, as the load

increased, positive pressures developed and friction dropped as the process moved towards partial

lubrication. Only positive pressure was measured for the convex wafers, even at low applied load, low

speed conditions. This wafer type was more affected by increasing speed, which moved the lubrication

phenomena move closer to fully hydrodynamic conditions with the associated reduced friction. No fully

hydrodynamic tribological conditions were observed in any of the range of conditions tested.

8

Submitted to the Journal of the Electrochemical Society References:

1. J. Levert, F. Mess, L. Grote, M. Dmytrychenko, L. Cook, and S. Danyluk, Slurry film thickness

measurements in float and semi-permeable and permeable polishing pad geometries, Proceedings of the International Tribology Conference; Yokahama, Japan (1995).

2. L. Shan, J. Levert, L. Meade, J. Tichy, and S. Danyluk, Journal of Tribology, 122(3), 539–543, (2000).

3. D. Bullen, A. Scarfo, A. Koch, D. Bramono, J. Coppeta, and L. Racz, Journal of the Electrochemical Society, 147(7), 2741–2743, (2000). 4. J. R. Coppeta, Ph.D. thesis, Tufts University, (1999). 5. J. Lu, J. Coppeta, C. Rogers, L. Racz, A. Philipossian, and F. B. Kaufman, Materials Research Society

Proceedings, 613:E1.2, (2000). 6. A.M. Scarfo, MS thesis, Tufts University, (2003).

9


Figure 1: CMP Polisher Setup

Wafer (45, 60, 75 rpm)

Vibration Isolation Table

Tabletop Polisher

Polisher Head

Pump

SlurrySolution

Variable Downforce (3-8 psi)

Polishing Pad(30-300 rpm)

Pressure Device

Friction Table

10


Plexiglas wafer

Adapters

1-way Valve

Magnet

Position Sensor

Transducers

Stabilization Pin

Computer

Slip Ring

Figure 2: Pressure Measurement Device

11


Figure 3: Wafer Tap Configuration

12


Figure 4: Wafer Shape Characterization (Type B pads are neither globally convex or

concave)

Concave-Type A Convex-Type C

Pad Pad

13


Figure 5: Normalized Pressure vs. Radius at 27.6 kPa

14

Chris Rogers

I would name them Concave, etc rather than type ABC (or both)


Figure 6: Normalized Pressure vs. Radius at 41.4 kPa

15


Figure 7: Net change in % of Load Supported by Fluid Film

16


Figure 8: % Reduction of COF

17


Table 1: Normalized Friction Measurements (Measured Divided By Applied Force) for all Conditions

27.6 kPa Downforce Wafer A Wafer B Wafer C

.35 m/s 31% 22% 17% .47 m/s 27% 18% 13% .58 m/s 24% 16% 10%


.35 m/s 24% 19% 18% .47 m/s 19% 14% 13% .58 m/s 17% 12% 10%

Table 2: Percentage of Downforce Exerted on Wafer Surface


.35 m/s -6% 22% 36% .47 m/s 3% 33% 51% .58 m/s 5% 38% 57%


.35 m/s 25% 32% 41% .47 m/s 31% 45% 55% .58 m/s 37% 51% 64%

18


19

in-situ measurement of pressure and friction during cmp of ... · measured friction coefficient...

Documents