Shedding New Light on Grinding and Polishing
One of the enabling technologies in the creation of microprocessors-devices that have revolutionized the world over the last several decades-is grinding and polishing through chemical-mechanical planarization (CMP). CMP allows the creation of high-density, multi-layered integrated circuits. Like all grinding and polishing processes, CMP takes advantage of the hardness of ceramics to produce a two-dimensional surface after a lithographic or deposition step. As the name suggests, there is a "chemical" in addition to the age-old "mechanical" aspect of this process, but the abrasive is essentially the same as the abrasives used to polish optics or ornamental stone. They consist of a ceramic material (usually silica, alumina or ceria) in particulate form, dispersed in a fluid. In the case of CMP, the abrasive is applied to a rotating pad, which is then placed in contact with the surface to be planarized.
Since the ceramic is in the form of millions of particles, the overall grinding/polishing properties of the abrasive will depend strongly on its particle size distribution (PSD). These properties, in turn, will determine what the grinding or polishing process will be. Particle size distributions are defined by a mean diameter and range. The larger both of these numbers are, the more surface material can be removed in a given amount of time.
Another important consideration is that dispersed particulate systems are inherently unstable. Eventually, the individual particles will come together to form large aggregates, and a significant concentration of aggregates will change the grinding/polishing properties of an abrasive. In the case of CMP, even small concentrations of aggregates will have a critical effect, since these large particles will scratch the planarized surface. Given the small distances between components in integrated circuits, these scratches or defects will render the resulting chips useless.
Clearly, CMP and other grinding/polishing processes can benefit from a particle sizing technique that can provide high-resolution/high-sensitivity PSDs. These distributions will not only allow a process engineer to more accurately determine whether a given abrasive dispersion is appropriate for achieving the desired surface properties, but also how to create a process (what grinding times, pad pressure, etc.) that will achieve this end. In addition, a technique that can detect small amounts of oversized particles or aggregates will allow an abrasive dispersion to be evaluated for its tendency to cause defects or scratches.
Measuring Particle Size DistributionA number of different techniques are available to measure particle size and particle size distribution.1 Two of the most common are laser diffraction and sedimentation.
Laser diffraction is based on the scattering of light from an ensemble of particles. The method makes use of the angular dependence of this scattered light on particle size to infer, through deconvolution algorithms and curve fitting, a particle size distribution. While this technique offers a wide dynamic range (which allows the technique to measure particles in the sub-micron and above-micron regimes) and fast measurement times, laser diffraction is a low-resolution/low-sensitivity method. As a result, the measured ranges are often much larger than in reality. This might lead an engineer into an incorrect determination of the grinding qualities of an abrasive dispersion, which might result in less than optimum process parameters. Furthermore, several studies have demonstrated that laser diffraction cannot be used to accurately measure small amounts of oversized particles.2,3
Sedimentation is another commonly used particle sizing technique. This technique is based on the relationship between the sedimentation velocities of particles in a liquid and the size of those particles as described by the Stokes Equation.4 These types of instruments have very good resolution but low sensitivity. While they can provide an engineer with useful information about the mean diameter and range of the abrasive dispersion, they cannot quantify the amount of oversized particles. Also, the measurement times tend to be quite long, and the dynamic range is limited.
A newer technique, called single particle optical sizing (SPOS), can measure high-resolution/high-sensitivity PSDs. This technique makes use of light obscuration to count and size particles one at a time. Because of this capability, SPOS can provide details about distributions that laser diffraction and sedimentation cannot.
An Illuminating TechniqueUnlike other particle measurement technologies, SPOS does not rely on fitting algorithms or ill-defined mathematical transformations. As seen in Figure 1, the technique involves passing particles through a small optical flow cell. A laser light source illuminates a thin section of the flow path, and the particles must pass through this detection zone on their way through the flow cell. As a particle passes through the view volume, the detectors will register a change in the intensity of the light. The response of the detectors is related to the size of the particle via a calibration curve.
When coupled with an appropriate fluidics system, SPOS can accurately characterize highly concentrated abrasive dispersions for mean diameter and range. It can provide a quantitative measure of oversized or aggregated particles even when present at parts per million (ppm) concentrations, and can be used to evaluate filters that remove the particles that will scratch a surface. It can also monitor shifts in particle size distribution due to usage and allow a process engineer to determine whether an abrasive has to be changed out.
While it has a limited dynamic range (0.5 to 2500 microns), SPOS provides excellent resolution and sensitivity in the above-micron range. However, it is not limited to testing materials composed of larger particles. SPOS can also provide useful information about primarily sub-micron distributions, such as the abrasive dispersions used for CMP.5 SPOS cannot provide information about the mean diameter and range of these materials, but it is still able to quantify the amount of large or oversized particles that are several standard deviations from the mean diameter.
As can be seen, the particle size distribution changed significantly. Instead of being a single, fairly narrow peak, the new distribution is bi-modal. The first peak, centered at 1.5 microns, primarily represents the abrasive particles that were reduced in size due to the polishing process. Also present are small particles removed from the polished surface. The second peak is from primary abrasive particles not reduced in size by the polishing process. Due to the polishing process, the mean diameter decreased to 6 microns, and the range widened to 0.5 to 30 microns.
It is clear from the graph that more than half the mass of abrasive material has been reduced in size enough to make the dispersion unusable for further polishing. It was the practice of the engineer in this case to send used abrasive to another facility to be recycled. The abrasive was classified using a hydrocyclone, which consists of a funnel where material is introduced tangentially. The centrifugal forces produced by the high-speed motion of the particles down the funnel separates particles in fractions based on size, which pass out different openings of the hydrocyclone. This was done to separate out the fine particles so that the larger ones could be used to form a new dispersion that might be reintroduced to the polishing process.
The yellow curve in Figure 2 represents the particle size distribution for the recovered abrasive dispersion. The hydrocyclone classification greatly reduced the concentration of fines; however, the mean diameter is still smaller (10 microns) than the fresh dispersion, and the range is wider. The standard deviation was 52%. This indicates that while the hydrocyclone helped, the dispersion still required further classification before being reintroduced to the process.
Distinguishing the DifferenceSPOS also has the ability to measure the concentration of oversized particles. This is illustrated in Figure 3a, which contains the results from abrasive dispersions of silica used in CMP. As detailed earlier, even small concentrations of large particles can damage chips and reduce the yield of in-spec devices. The PSDs in Figure 3 were obtained from two different abrasive dispersions. One (red curves) was determined to be "bad"-that is, visual inspection showed that it caused defects on the wafer surface that was polished. The other (blue) was not observed to cause defects.
To a light scattering particle size analyzer (Figure 3a), both samples appear quite similar in terms of mean diameter and range, with the so-called "bad" sample having a slightly lower mean diameter. Based on these results alone, a process engineer might be inclined to think that the "bad" slurry might cause less damage because it has a smaller mean diameter. However, this was not the case. When tested with SPOS (Figure 3b), which has the sensitivity to detect small amounts of large or oversized particles, the true difference is apparent. Clearly, the difference between the "bad" defect-causing dispersion and the "good" dispersion is found in the oversized particles and not in the mean diameter or range.
Optimized GrindingAs shown in the foregoing examples, knowing the particle size distribution of dispersed abrasive particles is crucial to achieving successful results from a grinding or polishing process. Accurate knowledge of the mean diameter, range and concentration of oversized particles is required for choosing the appropriate abrasive slurry and for determining the optimum process parameters. With the SPOS technique, engineers can analyze the particle size distribution of abrasive dispersions with high resolution and high sensitivity-and optimize their grinding and polishing processes.
For more information about the SPOS technique, contact Particle Sizing Systems, 8203 Kristel Circle, Port Richey, FL 34668; (727) 846-0866; fax (727) 846-0865; e-mail email@example.com ; or visit http://www.pssnicomp.com .