Correlation of FT-IR epitaxial thickness measurement
07/01/2000
Yiorgos Kostoulas, Gerhart Kneissl, Ian Kohl ADE Corporation, Westwood, Massachusetts
From an array of commercially available methods for measuring the thickness of epitaxial layers, including referee methods, a robust comparison has revealed that FT-IR spectroscopic analysis results in the most accurate measurements for the 0.3-25µm range. All other methods show significant "biases," but for thicker epi, the differences are less significant. In addition, the transition layer width and substrate doping numbers provided by the spectroscopic analysis-based technique agree well with SIMS results.
Fourier transform infrared (FT-IR) technology has been the method of choice for measuring silicon epitaxial layer thickness. There are currently two implementations of this technology: interferogram-based and spectroscopic analysis-based. The former was introduced in 1970, the latter a more recent introduction [1], first officially demonstrated at Semicon/West 1997.
Here, we compare the performance of the spectroscopic and interferogram-based techniques, as well as secondary ion mass spectrometry (SIMS) and spreading resistance probing (SRP), two other referee methods currently accepted by the industry. Although neither SRP nor SIMS represents an ASTM approved method to measure epitaxial film thickness, SIMS has generally been accepted as a practical referee method of choice.
Relevant theory
The currently available interferogram-based tools rely on the time domain to extract epitaxial layer (epi) thickness. An abrupt epi-substrate interface is assumed (i.e., no transition layer) and all the information is extracted from a single-beam measurement. The consequence of this single-beam approach is poor intermachine correlation as well as instrument drift. Because the method is relying on one optical beam for the measurement, any variation between tools, as well as changes in the light source or in environmental conditions, is visible in the results. In addition, as thinner epi layers become more prevalent, measuring those thicknesses using the interferogram-based technique becomes more challenging. For epi thinner than 3mm, the interferogram sidebursts begin to merge with the centerburst, and separating the two to extract the epi layer thickness requires the use of reference wafers. Because of all this, only one value, epi layer thickness, is extracted by the interferogram-based method, and this measurement tends to be unreliable for epi thickness <3µm.
Figure 2. Comparison of a) Lab B to Lab A and b) Lab C to Lab A.Click here to enlarge image
The spectroscopic, modeling-based approach is the most recent implementation of FT-IR-based film thickness metrology. Instead of relying only on the interferogram information, the instrument measures the reflectance (%R) of the sample over a wide spectral range. A synthetic reflectance spectrum is then calculated from first principles of optical physics and is fitted to the measured spectrum using three independent parameters- epi layer thickness, transition region thickness, and substrate carrier concentration. A least-squares fitting parameter is also calculated and serves as a measure of confidence in the fit.
This model-based approach takes into account that the epi layer substrate interface has a finite width. There is a smooth transition region in which the carrier concentration increases from that of the epi layer to that of the substrate. This continuously varying profile is sampled as a stack of layers each described by a constant carrier concentration, and therefore constant optical properties. This model-based software circumvents problems associated with interferogram-based systems when measuring thin epi layers. In addition, the reflectance spectrum by definition is a ratioed spectrum, making this a double beam measurement. This approach minimizes intermachine disagreement so much that it approaches single-machine precision.
System comparisons
We compared the performance of a spectroscopic tool, the EpiScan 1000 (dubbed Lab A [2]), with that of two interferogram-based tools (Lab B and Lab C). (The latter measurements were performed at the laboratories of MEMC.) In turn we compared the data gathered with SIMS and SRP.
We used a set of 24 p/p+ epi wafers with epi thickness ranging from 0.3 to 25mm, representing current production capabilities. After completing the FT-IR measurements, the wafers were cleaved along the notch bi-sector and each half was sent for SIMS and SRP analysis by third party laboratories.
SIMS analysis was performed by Charles Evans and Assoc., initially using oxygen ions on all samples. After discussion with the lab's scientists, we decided to use cesium ions for samples with epi thickness greater than 5mm, to achieve more accurate measurements.
Click here to enlarge image
The SRP measurements were performeed at Solecon Laboratories. The SRP data reduction employs a combination of the multilayer potential distribution approach [3] and the local slope approach [4].
The Lab A tool gave us three parameters: epi layer thickness, transition region thickness, and substrate carrier concentration (resistivity). The epitaxial film thickness is the distance from the surface of the epitaxial layer to that point on the transition layer profile where the carrier concentration reaches 50% of the substrate carrier concentration. The transition layer width was defined as the distance on the transition layer between the 10% and 90% points of the substrate carrier concentration.
The Lab B and Lab C tools were operated at their normal production use settings [5]. These tools, due to their interferogram-based, single-beam approach as discussed earlier, gave us only epi layer thickness.
Epi layer thickness
Figure 4. Comparison of substrate doping from Lab A and SRP vs. SIMS analysis. Data point representations: SRP: diamonds, and Lab A: squares.Click here to enlarge image
Figure 1 on p. 289 shows epi layer thickness measurements from the the SRP technique and the three different types of equipment as a function of the SIMS results. Figure 1a compares the performance of the various methods in the epi thickness range 0.3-3mm, Fig. 1b, 3-25mm. Lab A shows excellent agreement with SIMS over the whole range of epi thicknesses studied (no bias adjustment). The results of Lab B, C and SRP, on the other hand, diverge significantly from the SIMS values. For thicker epi, the differences are relatively small and only SRP seems to diverge somewhat for epi thicknesses of 14µm.
In addition, we show the correlation results between Lab B, Lab C, and Lab A (Fig. 2a and 2b). The correlation between Lab B and A shows an offset (bias) of about 180nm relative to Lab A. This offset drops to 90nm for Lab C.
Transition region thickness
To analyze transition region thickness, we compared the performance of Lab A and SRP with SIMS (Fig. 3). Recall that Lab B and Lab C are interferogram-based tools and therefore do not provide measurements of the transition layer width. Figure 3 shows there is excellent agreement between the values obtained using the analysis-based FT-IR method of Lab A and SIMS. SRP values also correlate well with SIMS but are more broadly distributed around the parity line. This clearly demonstrates the need to use a graded profile to describe the epi-substrate interface.
Substrate doping
The analysis-based FT-IR technique is also the only one that provides information on substrate doping. Figure 3 show a comparions of Lab A and SRP measurements as a function of SIMS. It can be seen that Lab A correlates well with SIMS.
Conclusion
Our comparisons of different commercially available FT-IR epitaxial film thickness tools, with each other as well as with the broadly accepted de facto referee methods of SIMS and SRP (Table 1), should prove valuable to epi process engineers. We found that for thin epi — 0.3 to 3.0µm — a spectroscopic tool is the only one that agrees completely with SIMS. All other methods show significant "biases." For thicker epi, the differences are less significant. In addition, the transition layer width and substrate doping numbers provided by the spectroscopic, analysis-based technique agree well with SIMS results. The correlation results are summarized in Table 1.
Acknowledgments
We thank Dr. Robert Standley of MEMC, St. Peters, MO, for supplying the special set of epi wafers used in our tests.
References
- IBM Instrument Division introduced a model based FT-IR epitaxial film thickness gauge in the early 1980s. The tool was withdrawn from the market when IBM closed the Instrument Division.
- EpiScan 1000, the only spectroscopic tool available on the market today.
- P.A. Schumann Jr., E.E. Gardner, J. Electrochem. Soc., Vol. 116, pp. 87-91, 1969.
- D. Dickey, J. Ehrstein, NBS Special Publication 400-48, May 1979.
- One of the difficulties in comparing thickness values obtained by different FT-IR tools is the fact that interferometric-based tools allow the operator to chose between different "peak picking programs," each program producing different values. After some discussion it was agreed that the MEMC lab would use the program normally employed in routine production applications.
Yiorgos Kostoulas received his PhD in semiconductor physics from the University of Rochester. He has six years of experience in semiconductor characterization. Kostoulas is a product manager with ADE Corp., 77 Rowe St., Newton, MA 02466; ph 617/831-8052, fax 617/243-4452, e-mail [email protected].
Gerhart Kneissl received his PhD in mechanical engineering from Oklahoma State University. Kneissl is marketing manager for thin film products with ADE Corp.
Ian Kohl received his MS in physics from Brigham Young University. Kohl is an applications engineer with ADE Corp.
Figure 2. Comparison of a) Lab B to Lab A and b) Lab C to Lab A.
This model-based approach takes into account that the epi layer substrate interface has a finite width. There is a smooth transition region in which the carrier concentration increases from that of the epi layer to that of the substrate. This continuously varying profile is sampled as a stack of layers each described by a constant carrier concentration, and therefore constant optical properties. This model-based software circumvents problems associated with interferogram-based systems when measuring thin epi layers. In addition, the reflectance spectrum by definition is a ratioed spectrum, making this a double beam measurement. This approach minimizes intermachine disagreement so much that it approaches single-machine precision.
SIMS analysis was performed by Charles Evans and Assoc., initially using oxygen ions on all samples. After discussion with the lab's scientists, we decided to use cesium ions for samples with epi thickness greater than 5mm, to achieve more accurate measurements.
The Lab A tool gave us three parameters: epi layer thickness, transition region thickness, and substrate carrier concentration (resistivity). The epitaxial film thickness is the distance from the surface of the epitaxial layer to that point on the transition layer profile where the carrier concentration reaches 50% of the substrate carrier concentration. The transition layer width was defined as the distance on the transition layer between the 10% and 90% points of the substrate carrier concentration.
Figure 4. Comparison of substrate doping from Lab A and SRP vs. SIMS analysis. Data point representations: SRP: diamonds, and Lab A: squares.