High-Resolution profilometry for CMP process control
06/01/1997
High-resolution profilometry for CMP process control
Jason Schneir, Renee Jobe, Tencor Instruments, Milpitas, California
Vincent W. Tsai, University of Maryland, College Park, Maryland
Chemical-mechanical polishing (CMP) process cells commonly use profilometry to measure the post-CMP planarization of the wafer. However, as the feature size of ultra large-scale integration technology decreases, it has become increasingly difficult to resolve individual surface features using a profiler. Here we present a new instrument, a high-resolution profiler (HRP), which can be used for surface metrology in the spatial wavelength range of 30 mm to 10 nm. HRP can, for the first time, measure the wafer surface on the macroscopic and microscopic scale with the same instrument.
Developing damascene process modules and bringing them to full-scale manufacturing is a challenging task. Temporal variation in the CMP process results from equipment and consumable drift as well as other disturbances that occur as wafers are processed. Equipment or process limitations result in spatial variations across each wafer, and variations within each die are affected by complex pattern dependencies. As we will see, these process variations can lead to significant parametric yield loss if process parameters are not brought under control.
Challenges in metal CMP
The process engineer uses models to bring metal CMP under control and increase process margins. Yield learning depends on developing models that relate process parameters to product wafers. Unfortunately, it is difficult to develop adequate models for metal CMP. Metal polishing involves complex surface chemistry and complicated interactions between the various polishing parameters. The fundamental science of metal polishing is not well understood and most company information is proprietary. Each company must develop its own knowledge base on metal CMP. Process margins cannot be defined until the effects of process variations are understood.
Process variations in metal CMP can lead to many different failure modes. For example, most metal-polishing processes result in recessed metal features. Changes in the process, such as the pH level of the slurry, polishing speed, or conditioning of the pads, would change the size of plug recess. If the depth of recessed metal plugs is too great, it will degrade the electrical connections between the vias and metal lines used as interconnects. Generally, plugs with 0.25-?m geometry must have a recess <50 nm. If the recess size exceeds 50 nm, the interconnect deposition will not be able to fill the deeply recessed vias completely. However, even if the recess is <50 nm, it can cause problems: it is difficult to clean the polishing slurry out of recessed areas. Slurry adhering to the wafer can cause contamination problems in subsequent process modules. Thus, parametric variations in the CMP process can lead to slurry contamination in subsequent process modules.
Erosion is another common cause of problems. Areas with a high density of metal features polish faster due to the enhanced lateral polishing rate of the interlayer dielectric (ILD) at exposed edges. Erosion adversely affects the planarization of the wafer and eats into the DOF budget, reducing the process latitude and possibly the yield for subsequent lithography.
During damascene CMP, some process variations can lead to failure modes that are difficult to trace back to the source. For example, the stepper alignment marks may be polished asymmetrically by the CMP process (Fig. 1). The stepper optical system then incorrectly interprets the asymmetric shape of a mark as a shift in its lateral position. The resulting parametric overlay error combines with random overlay variations to produce difficult-to-trace yield loss.
 |
Figure 1. Asymmetric polishing of stepper alignment marks and overlay marks can cause overlay error. This can result in misalignment between the current mask level and the subsequent level.
Planarization metrology requirements for metal CMP
The challenges of metal CMP can best be met by carefully characterizing process margins during process development and implementing an efficient process-control strategy during full-scale manufacturing. The metrology tool used for these measurements must have, at a minimum, the following performance characteristics:
A scan length greater than 50 mm. One key metric for metal CMP performance is global planarity. Erosion and dishing effects cause pattern-dependent variations in global planarity on the length scale of millimeters (within die). Process and consumable drift and/or limitations can lead to variations in flatness across the wafer. The metrology system must be able to detect and characterize these variations.
Ability to image 0.25-?m features. As discussed previously, the recessing of tungsten plugs caused by metal CMP must be kept under control. The metrology system must be capable of making high-resolution measurements on the recess and dishing of interconnects in dual damascene processes.
Repeatability of better than 1 nm ( = 3 s, where s is the standard deviation) on depth measurements. Typically, plug recess must be controlled to within 10 nm; the metrology tool must have a repeatability of 1 nm on plug-recess measurements.
The new HRP was developed to meet these performance requirements.
Micro- and macro-imaging
Currently, two types of probe instruments are used for CMP metrology:
1. Stylus profilometers, developed in 1936, are commonly used in the CMP process sector, and
2. Atomic force microscopes (AFMs), invented in 1985, are commonly used in analytical analysis labs.
Stylus profilometers are rugged instruments based on a technology that has been refined over many years. They have vertical resolutions of better than 0.1 nm (or 1 ?) and lateral resolutions of ~.03 ?m in the x direction and 1 ?m in the y direction. The AFM was invented to overcome the lateral resolution limitations of the stylus profilometer. It has ~1-nm lateral resolution in both the x and y axes, but this improvement reduced the scan length. Most commercial AFMs have a maximum scan length of less than 100 ?m. (The maximum scan length demonstrated using an AFM is 500 ?m [1].)
HRP provides long travel (200 mm) and nanometer lateral resolution. Previously, completing a surface study on both the microscopic and macroscopic scale required separate AFM and profiler studies, and it was impossible to correlate the global surface structure with nanometer-scale AFM results. HRP technology overcomes this limitation.
The six HRP images at different magnifications in Fig. 2 demonstrate the wide scan and resolution of the HRP. Each image is obtained by scanning a sharp diamond stylus over the surface in an x-y raster pattern. Fig. 2a shows a 50 ? 50 mm view of an intentionally underpolished oxide film. Fig. 2b shows a new image that zooms in on a 5 ? 5 mm section of the wafer. Each subsequent image zooms on a section of the previous image with a factor of 10 higher magnification. Finally, we zoom in on polishing debris on the surface.
 |
Figure 2. Six views of a CMP surface imaged using HRP. The top three images can be obtained with a profiler, while the bottom three images previously required an AFM. The HRP is the only instrument that can provide all six images (five orders of magnitude).
Figure 3 shows a HRP image of 0.25-?m geometry tungsten plugs after CMP. The development of HRP provides the resolution needed without sacrificing the long-scan capability of a profiler. Figure 4 shows the spatial wavelength range resolvable corresponding to an HRP, AFM, profiler, optical profiler, and phase-measuring interferometers. The HRP has the ability to image the surface over a wider spatial wavelength range than any other instrument.
 |
Figure 3. HRP image of 0.25-?m tungsten plugs.
 |
Figure 4. A comparison of the spatial frequency capability of a number of different metrology tools. Optical profilometer tools can produce erroneous results on CMP wafers due to reflections from the underlying films.
Sensor design
The HRP sensor (Fig. 5) consists of an electromagnet, a flexure hinge, and a capacitance sensor. As the stylus scans over the contours of the surface, it moves up and down. The flexure hinge constrains the stylus arm to pivot about the hinge in a seesaw motion. The capacitance sensor measures the height of the stylus, which is proportional to the separation of the capacitor plates, d. If we ignore the variations in the electric field near the edges of the plate, we find a simple relationship between the distance between the plates, d, and the capacitance C:
 |
Figure 5. The HRP sensor allows an Al surface to be imaged nondestructively.
Therefore, by measuring the change in capacitance of the plates, we can determine the change in spacing, d, and the height of the surface, s. The HRP capacitance sensor can measure the surface height with two range settings. For a range of 13 ?m, the resolution is better than 0.01 nm; for a range of 65 ?m, the resolution is better than 0.1 nm.
Obtaining high-resolution data requires a sharp stylus that will degrade and/or damage the surface if the force is too large. We can estimate the effect of using a sharp tip by modeling the contact as a spherical ball pressing on a flat surface. This model of the interaction, known as a Hertzian contact [2], assumes a spherical tip and ignores frictional effects. Using the Hertzian model, we find that the maximum force which can be applied without damaging the surface is [2]:
 |
where
R= the radius of the tip
sst = the ultimate strength of the surface
g1 = the Poission ratio for the surface material
E* = the equivalent elastic modulus
If the radius (R) of the tip is small, the force (f) must be kept small. The HRP provides a constant force independent of stylus position. This force is similar to those used in contact, tapping, or intermittent contact AFM. The electromagnet shown in Fig. 5 maintains the force at a very small, constant value (as low as 0.05 mg or 4.9 ? 10-7 N).
The easiest way to explain its operation principles is to carry out a two-part analysis. First, assume that the pivot does not have any stiffness and, consequently, has no spring force associated with it. We can then apply a constant force using the electromagnet and determine the surface height with the capacitance sensor.
In reality, the pivot does have some stiffness. When a new stylus is placed in the HRP, the electromagnet swings the pivot through its whole range. The corresponding force vs. distance curve for the pivot is recorded. A digital signal processor-based control system is then used to cancel out the spring force of the pivot so that the force is constant at all times.
The HRP sensor allows a gold surface to be imaged nondestructively. For example, we imaged a gold surface at a force of 0.05 mg (4.9 ? 10-7 N) using HRP, followed by scanning electron microscope (SEM) imaging. No scratching of the surface was seen on the SEM. HRP imaging of hand-baked photoresists also resulted in no observable scratches under the SEM.
Dual-stage technology
The HRP`s micro- and macro-imaging capability requires the ability to position the stylus with nanometer resolution and a range of 200 mm. The dual-stage system accomplishes this task (Fig. 6). The sample stage, used for macro-imaging, is a mechanical stage with a range of 200 mm and a repeatability of 1 ?m (1 s). The drive mechanism of the stage is based on a mechanical screw with a pitch of 1 mm. The carriage slides on a glass reference flat that constrains the motion of the sample stage to a plane. The out-of-plane motion of the sample stage is less than 1 nm.
 |
Figure 6. Dual-stage technology enables micro- and macro-imaging.
The sensor stage, used for micro-imaging, has a range of 75 ?m and 1-nm resolution. It consists of piezoelectrics mounted in a mechanical frame. As the translators expand, the mechanical frame expands as well. The mechanical frame is designed to remove undesirable bending motions of the piezoelectrics and only allow the desired x-y rectilinear expansion. As we will see in the next section, the flexure mechanism keeps the piezoelectric bending motions from degrading the HRP`s repeatability. The piezoelectric translators in the sensor stage have hysterisis and creep that, if not eliminated, would degrade the repeatability of the HRP. The sensor stage contains two capacitance sensors, both with resolutions of 1 nm, to determine the x and y positions of the stage. Using the information from the capacitance sensors, a feedback loop eliminates hysterisis and creep effects.
Measurement repeatability
High-resolution images such as Fig. 2 are valuable in an R & D environment. However, in a semiconductor fab, numbers are used to characterize and control manufacturing processes. Thus, the measurement repeatability of HRP is of paramount importance.
The Jorgensen procedure has been developed [3] for determining the x-, y-, and z- repeatabilities of AFMs. We used it to determine the repeatability of the HRP from the standard deviation of the x-pitch, y-pitch, or z-height measurements. A VLSI surface topography artifact (STS-1800) [4] was first imaged. The sample was then removed from the HRP, inserted back in the HRP system, and re-imaged. This procedure was repeated five times. The stylus of the HRP was then changed and the sample was imaged five more times, with the sample removed between each measurement. The same measurement protocol was then performed using a commercially available AFM and the data was analyzed using an algorithm developed by Jorgensen [3].
Figure 7 shows the measurement repeatability of a commercially available AFM (Fig. 7a) and the HRP (Fig. 7b). The respective x-, y-, and z- repeatabilities were determined to be 23 nm, 11 nm, and 3 nm (or 1 s) for the AFM, and 5 nm, 6 nm, and 0.3 nm (or 1 s) for the HRP. Removing the sample between measurements and changing the stylus provides a measurement (dynamic repeatability), which reflects how the tool will perform in a manufacturing environment. In our study, we found that the z-repeatability of the HRP is 10 times smaller than AFM, due primarily to two design differences:
1. The HRP uses capacitance sensors in the x, y, and z directions to avoid hysterisis and creep, while the AFM used for our test does not use capacitance sensors; and
2. The AFM cantilevers can twist and bend during imaging, while the HRP force sensor is not susceptible to this type of distortion.
 |
Figure 7. a) Repeatability of a commercial AFM on a 1.8-?m pitch and 180-nm deep VLSI surface topography standard. The x and y repeatabilities are plotted on the left-hand axis and the z repeatability is plotted on the right-hand axis. The repeatability is 23-nm (x), 13-nm (y), and 3.0-nm (1 s) (z). b) Results using a HRP. The repeatability is 5-nm (x), 6-nm (y), and 0.3-nm (1 s) (z). The repeatability is much better than the AFM repeatability.
CMP process control strategies - correlation scans
We used HRP to measure the post-CMP plug recess on various tungsten plug test patterns. We found that the plug recess depends on the pitch of the pattern (see table). That is, the recess of a tungsten plug is influenced not only by the size of the plug and the polishing, but also by how many other plugs are nearby. This pattern density effect occurs in all types of CMP. Understanding its effects on the process is essential to developing the models needed for control and optimization of metal polishing.
A complete picture of the mechanism responsible for plug recess would include the polishing rates of the ILD, tungsten, and adhesion layer (e.g., Ti/TiN). However, to a very rough approximation, the recess depth depends on the rate at which the tungsten polishes relative to the rate at which the ILD polishes. Since the tungsten polishes faster, its pattern becomes recessed with respect to the ILD. The tungsten area continues to recede until it is no longer in good contact with the pad. With the tungsten recessed, the pressure on the ILD is higher relative to unpatterned regions of the wafer. The increased pressure increases the polishing rate, causing more rapid polishing of the patterned area. In metal CMP, the ILD area with higher-feature density polishes more quickly. This effect is known as erosion.
For isolated plugs, the ILD polishes slower than the tungsten plugs and the plug recess is large. In areas with plugs packed close together, the ILD polishes more rapidly due to erosion, and the recess is smaller.
The dependence of erosion and plug recess on pattern density is important not only for process control, but also from a circuit-design standpoint. HRP can measure both plug recess and erosion by using micro- and macro-imaging.
Figure 8 shows HRP images of two adjacent tungsten plug test patterns. The two high-resolution HRP images reveal that the average height and pitch of the tungsten plugs on the left are 9 nm and 3.8 ?m, while the height and pitch for the pattern on the right is 10 nm and 4.8 ?m. Since the pitch of the pattern on the left is smaller, we expect that, due to erosion, this pattern will polish faster than the pattern on the right. The 8-mm long HRP trace reveals that the pattern on the left is 31 nm lower than the pattern on the right. The combination of high-resolution images and a long scan is called a correlation scan. A correlation scan provides the height of each plug over the local ILD (plug protrusion) and the relative height of the two micro-images (the erosion).
 |
Figure 8. A correlation scan provides the height of the plugs and erosion.
Correlation scans provide a powerful mechanism to study pattern-density effects in metal CMP. Figure 9 shows another example of a correlation scan. The combination of the micro-images and the long scan gives the plug recess in various regions and the relative height of each micro-image.
 |
Figure 9. The combination of micro- and macro-imaging provides the information necessary for CMP process control.
Automation
The HRP has the cassette-to-cassette automation common in profilers used in the semiconductor industry, including pattern recognition, die grid coordinates, sequences, and automatic wafer handling. In addition, an automatic tungsten-plug-finding algorithm allows high-resolution traces to be quickly taken directly through the center of 0.25-?m tungsten plugs. It can locate a feature with a positioning repeatability of 1 ?m. To take a scan automatically through the center of a tungsten plug, the stylus searches for the plug and short scans spaced by an appropriate interval (e.g., 0.15 ?m for a 0.25-?m plug) are obtained, until one of the scans crosses the plug. The center of the plug is then determined and a scan is taken through the center of the plug.
Summary
HRP technology addresses the complex requirements of metal CMP metrology. It can provide scans up to 200 mm; image features down to 0.25 ?m; measurement repeatability of 5 nm (x), 6 nm (y), and 0.3 nm (1 s)(z); nondestructive imaging; and correlation scans, offering IC manufacturers a single system to control metal CMP parameters. Additionally, IC manufacturers will be able to correlate micro- and macro-surface analysis to obtain comprehensive information on the global surface structure. The HRP provides the information needed for process development and has the automation and robustness needed for manufacturing. This ability to use the same tool in both environments will expedite the transfer of advanced CMP applications to full-scale manufacturing.
Acknowledgment
The High Resolution Profiler (HRP) is a trademarked product of Tencor Instruments. The authors appreciate the input and encouragement of Mansour Moinpour of Intel Corp. They also thank their colleagues A. Samsavar and W. Sze at Tencor Instruments, J. Chu and M. Maxim at Intel Corp., and their collaborator D. Hetherington of Sandia National Laboratory.
References
1. J. Fu, R.D. Young, T.V. Vorburger, Rev. Sci. Inst., Vol. 63, p. 2200, 1992.
2. S.P. Timoshenko, J.N. Goodier, Theory of Elasticity, 3d. ed., McGraw-Hill Inc., 1987.
3. G. Barbatro et al., "Scanning Tunneling Microscopy Methods for Roughness and Micro Hardness Measurements," Synthesis report on BCR project 3423/1/0/184/4/91-BCR-DK(30), 1994.
4. STS-1800, VLSI Standards Co.
Jason Schneir received his PhD degree in physics in 1988. He has worked on applications of scanned probe microscopy for 11 years and has 39 publications in the field. He is a technical product manager for the High Resolution Profiler at Tencor Instruments. Metrology Divison, One Technology Drive, Milpitas, CA 95035; 408/571-3000, fax 408/571-3030.
Renee Jobe received her BS degree in physics from California State Polytechnique, Pomona, 1989. She is a senior applications engineer in the Metrology Division at Tencor Instruments.
Vincent W. Tsai received his MS degree in materials engineering from the University of Maryland, College Park. He is a PhD candidate in materials engineering at the University of Maryland, College Park, and a guest researcher at the National Institute of Standards and Technology (NIST). He has researched the use of scanning probe microscopes for metrology.