Broadband spectrophotometry for wafer-level packaging
BY John C. Lam
A growing number of semiconductor manufacturers use wafer bumping packaging technology to pack chips more closely together for such applications as cell phones, where space is at a premium. The bump bonding technique requires ultra-thick photoresists to define bond size and location. These photoresist layers typically are in the range of 50 to 100 µm (or more) thick, which is substantially thicker than resists used in IC manufacturing.
Because of the high costs associated with defects and subsequent rework, accurate characterization of photoresist layers is critical to optimize yields. Currently, manufacturers use either a contact method such as profilometry, or a cross-sectional method such as scanning electron microscopy (SEM), to measure these ultra-thick photoresists. Standard optical characterization methods would be an option, but they are limited to a layer thickness of approximately ≤30 µm. The primary difficulty with conventional optical methods such as ellipsometry/ reflectometry is that these methods generally do not use the advanced algorithms required to resolve the reflectance spectra typical of ultra-thick layers.
The Broadband Spectrophotometric Method
Broadband spectrophotometry is a materials characterization method capable of overcoming the limitations of conventional optical methods by using a unique, all-reflective optical design in addition to proprietary modeling algorithms.
Optical characterization of thin films involves determining the film's thickness, t; index of refraction, n; and extinction coefficient, k. The latter two parameters frequently are described as optical constants, but the word “constant” is misleading because the values of n and k actually depend on the wavelength of light used to make the measurement. The index of refraction describes how the path of a beam of light deviates (refracts) as it passes through a material — the greater the value of n, the greater the refraction. The extinction coefficient relates to the absorption of light — the value of k increases as more light is absorbed by a material. Transparent materials have an extinction coefficient of zero in the visible wavelength spectrum. The film's energy band gap, Eg, is another useful parameter: it defines the minimum energy needed to induce an electron transition from valence to conduction band by photon absorption.
 Figure 1. Configurations of the light source and detector used in broadband spectrophotometry. |
null
As previously mentioned, an appropriate algorithm, or physical model, is essential to derive useful information from an optical measurement. To be able to characterize a thin film by optical means, it is necessary to develop an algorithm that expresses reflectance in terms of other properties including film thickness and the n and k spectra. The Forouhi-Bloomer equations1–3 constitute a physical model for n and k that is applicable to a wide range of semiconductor, dielectric and thin metal films deposited on opaque or transparent substrates. The model is valid over the deep ultraviolet (DUV) to near infrared (NIR) wavelength range, and is simpler than other models that have been used to describe n and k.
Characterizing Ultra-thick Films
In the following example, two wafers coated with very thick resist were characterized using the broadband spectrophotometric method. The system was used not only to measure the thickness of the resist layers, but also to determine simultaneously the n and k spectra of the material.
The samples were provided by a leading photoresist manufacturer. The first sample consisted of a patterned sample of ~70 µm photoresist on a silicon substrate. The second sample was unpatterned with ~100 µm of photoresist on silicon substrate. Pattern-recognition software was used to measure the first sample, with a UV filter in place to demonstrate UV-protection capabilities for photosensitive materials. The second sample was measured as is. Broadband reflectance data were acquired over the 190 to 1,000 nm spectral range. The measurement instrument used both deuterium and tungsten light sources, in combination with a fixed holographic diffraction grating and linear Si photodiode array (Figure 1). A bare Si wafer calibrates the spectral output of the light sources prior to undertaking reflectance measurements. Reflectance data are collected in a near-normal incident geometry.
Ten consecutive 49-point mapping scans were run on each wafer. The wafers were run consecutively to ensure that the samples were removed from the stage between trials. In this manner, the positional inaccuracy of the stage (~1.25 µm) was included in the measurement. Reproducibility was calculated as the standard deviation (STD) of the means of the 10 scans. As can be seen from the data in the table, the broadband spectrophotometric method produced excellent results for both the 70 and 100 µm films, with reproducibility in the 0.02 to 0.04 percent range.
null
The top graph shown in Figure 2 presents the results of reflectance measurements performed on the ~70 µm photoresist sample (red line). In this case, a UV filter was used so that reflectance in wavelengths 450 to 1000 nm would be recorded. Calculated reflectance spectra (green line) also are presented in the graph and are found to agree well with the experimental results. The n and k spectra, presented in the lower graph in Figure 2, were derived simultaneously with the reflectance spectra. Figure 3 presents 49-point wafer maps for the 70 µm sample, which is typical of the 10 trials.
 Figure 3. A map of 49-point thickness measurements from the ~70 µm-thick photoresist sample. |
null
A trial also was made to compare the performance of the broadband spectrophotometric method with profilometry in the characterization of photoresists of thicknesses up to
60 µm. Figure 4 shows that the results correlated quite well, indicating that the optical technique can provide equivalent data to the contact method, with the added benefits of being fast and nondestructive.
 Figure 4. Broadband spectrophotometry and profilometry measurements for photoresists up to 60 µm thick show good agreement. |
null
Measurement Speed and Throughput
Measurements made using broadband spectrophotometry, such as those previously presented, typically require ≤2 to 3 second/point. Each of the 49-point trial measurements shown here took ~2 minutes, not including the time to load and unload the sample. Therefore, throughput depends on the number of points measured per wafer. Because the measurement time itself is so short, throughput actually is limited only by operator speed.
Although the data presented here covered only ultra-thick photoresists to 100 µm, the nature of the diffraction grating in the detector will allow for thicker samples. This is because in wavelength space, the method described here samples information in increments of 1 nm, but in energy space, information is sampled in proportion to inverse wavelength (1/l). The extra information obtained differs slightly from the normal equal sampled interval, and ambiguity is removed from any aliasing.
Broadband spectrophotometry, with the appropriate physical model and data analysis methods, is a technology that performs equally well under standalone or integrated conditions, and it has the potential to bring substantial cost savings in the form of increased fabrication productivity and sustainable quality control.
References
1A.R. Forouhi, I. Bloomer, “Optical Dispersion Relations for Amorphous Semiconductors and Amorphous Dielectrics,” Phys. Rev. B, 34, pp. 7018–7025, 1986.
2A.R. Forouhi, I. Bloomer, “Optical Properties of Crystalline Semiconductors and Dielectrics,” Phys. Rev. B, 38, pp. 1865–1874, 1988.
3A.R. Forouhi, I. Bloomer, US Patent No. 4,905,170, 1990.
John C. Lam, senior applications engineer, may be contacted at n&k Technology Inc., 4051 Burton Dr., Santa Clara, CA 95054; (408) 982-0840; Fax: (408) 982-0252; E-mail: [email protected].