Mixture Analysis Made Easier: Trace Impurity Identification in Photoresist Developer Solutions Using ATR-IR Spectroscopy and SIMPLISMA Michel Hachey, Michael Boruta Advanced Chemistry Development, Inc. Toronto, ON, Canada www.acdlabs.com Steve Hill WaferTech, a TSMC company Camas, WA, USA www.wafertech.com Outline The usefulness of a simple-to-use self-modeling mixture analysis algorithm for identifying batchto-batch raw material variations and trace product contaminations is shown in the context of a photoresist developer solution. SIMPLISMA permits component detection and extraction of ppm levels of impurities even in the presence of strongly absorbing constituents for spectra of mixtures acquired by an ATR-IR instrument. Introduction Due to their acquisition speed and ease-of-use, techniques such as ATR-IR, UV-Vis, or Raman are often found at the front lines of QA/QC processes in the advanced materials manufacturing industry. However, spectra obtained using traditional molecular spectroscopy techniques can be especially challenging to interpret when a sample is a mixture of substances. It can be particularly difficult, for example, to detect and identify trace impurities in mixtures against the background of strongly absorbing primary components. Therefore analysis of intricate samples generally calls for the prior extraction or separation of isolated sample components if feasible in conjunction with the application of molecular spectroscopy techniques to identify and semiquantify each chemical entity found in the mixture. However, for solid state materials, inorganic mixtures, highly reactive substances, or organic reaction intermediates, there is often no physical or chemical means of isolating components. In such cases, chemometric software algorithms can offer the only practical means of extracting pure component spectra and concentration profile information necessary to facilitate mixture interpretation of molecular spectra. In this application note, we demonstrate how the SIMPLISMA [1] self-modeling software algorithm for mixture analysis can help speed up the interpretation of mixtures by extracting individual spectral components and concentration profiles from a mixture series without having to resort to chromatographic or physical separation methods. SIMPLISMA was used to help extract and identify trace impurities and contaminants in an ATR-IR spectral data set acquired from solutions of photoresist developer used in the manufacture of silicon wafers.
Algorithm The SIMPLe-to-use Interactive Self-modeling Mixture Analysis (SIMPLISMA) algorithm can provide an easy way to deformulate spectra of complex mixture series into their isolated pure components without having to use conventional chemical or physical extraction methods. This patented algorithm was implemented under license from Kodak in ACD/UV-IR Manager [], which provides a ready-to-use interface designed to make the algorithm more readily accessible to chemists and spectroscopists as a routine analysis tool. In a nutshell, SIMPLISMA relies on finding pure variables to mathematically resolve complex spectral matrices. A pure variable is one where only one component of the mixture shows a response and the other components remain silent. SIMPLISMA essentially uses a function for identifying a pure variable for each pure component in the mixture, and collects the intensity responses of the variables at each of these locations. These pure component intensities are used as concentration estimates to help resolve the spectral data matrix by least squares regression. This provides us with pure component spectra and concentration profiles, respectively, completing the curve resolution process. Note For more accurate descriptions and details on how the algorithm works, please refer to the literature [1,3-5]. Preprocesing Requirements SIMPLISMA makes certain assumptions about the data set that can impact whether it can be applied to certain problems. The first assumption is that the system does indeed contain some pure variables, which unfortunately is not always the case. Second, it also assumes that the additivity rule generally applies to the sample set. Third, it requires that the target component s sample composition ratio (e.g., concentration) must show some change for the algorithm to detect its presence. Enough systems pass those criteria to sufficient degrees for SIMPLISMA to have found a very wide base of application []. Data not meeting the above criteria can sometimes be processed in a way to remove, or at least suppress or masks undesirable characteristics. For example, since non-additive responses are often localized to certain spectral regions, it is often expedient to simply cut-off the problem areas before attempting the analysis on the remaining data. As another example, the use of the second derivative can help resolve overlapping component bands or sloping backgrounds and thus makes it easier to find a reasonably pure variable within a complex data set. Last but not least, since the lack of pure variables is often associated with the presence of a non-zero background, background subtraction methods can go a long way towards improving the suitability of a data set for SIMPLISMA analysis. ACD/UV-IR Manager [] provides Group Baseline correction tools and capabilities to subtract the minimum spectral plot from a series that are useful for correcting background problems prior to SIMPLISMA analysis. If complex spectral behavior does not allow for an unambiguous selection of nodes for a Group Baseline correction, then subtracting the minimum spectral plot from the series is usually considered. ACD/UV-IR Manager allows the extracted minplot from a spectral series to be subtracted from the entire series in just one operation.
Experimental All experiments were performed with an ATR infrared instrument using neat samples. In each case, the diamond artifacts between 1700 and 300 were zapped from the spectral series. Since substantial non-linear constant backgrounds needlessly complicate human interpretation and are detrimental to the SIMPLISMA curve resolution algorithms, these were removed by extracting the minimum plot for the series and subtracting it from every spectrum therein. Detecting Low Level Impurities in Aqueous Photoresist Developer Concentration of the aqueous photoresist developer determines the resolution of the pattern on a wafer, and it therefore needs to be tightly controlled. Aqueous photoresist developers are normally very stable systems, being composed of high-purity tetramethylammonium (TMA) hydroxide blended with water at a target concentration precise to within ±0.00%. If undesirable variation occurs in the development rate, these are normally attributed to either a shift in concentration, or impurities and contaminations by method. Can we use SIMPLISMA on ATR-IR measurements to help extract and identify low level organic impurities not currently reported in the existing analysis reports? [x 10 - ] Infrared Absorbance 0 - - - 000 5500 5000 500 000 3500 3000 500 000 1500 1000 Wavenumber (cm-1) Figure 1: ATR-IR measurements with water as reference of aqueous photoresist developer samples. Samples from 1 lots of photoresist developer solution were taken at several process locations and measured by ATR-IR with water as a reference. The resulting spectral series is shown in Figure 1. This spectral data needs some pretreatment before it can be analyzed by SIMPLISMA. First, the region from 000 to 000 cm -1 was truncated out of the data set because it behaves like noise that would needlessly hinder our analysis. Second, the minimum plot for the series was extracted and subtracted from the series in order to remove the constant background, and, more importantly, to remove negative absorbance areas caused by the use of a pure water reference (SIMPLISMA cannot process spectra with negative areas). Finally, the diamond band artifacts between 1800 and 300 cm -1 were zapped because they are irrelevant to the mixture composition. After all this pre-processing, the spectral series now takes a more readily interpretable shape as shown in Figure. 3
[x 10-3 ] 8 7 Infrared Absorbance 5 3 1 000 3800 300 300 300 3000 800 00 00 00 000 1800 100 100 100 1000 800 Wavenumber (cm-1) Figure : ATR-IR of aqueous photoresist developer samples after pretreatment. Analysis with SIMPLISMA helped resolve three primary components as seen in Figure 3. It is believed that the top spectrum (in red) corresponds to TMA and TMA carbonate, which have very similar absorption profiles that might have prevented SIMPLISMA from resolving them from one another. The TMA carbonate likely accounts for levels less than 0.01% (<100 ppm). The sensitivity of SIMPLISMA to relative composition variation is clearly evident by a relatively weak TMA profile that represents a true and stable composition over 100 larger than the TMA-CO3. The middle spectrum (in green) is attributed to water with signs of protein contaminants due to bacterial growth in the water bottle used for cleaning. This contaminant is introduced during cleaning between samples. The bottom spectrum (in blue) seems to contain some water contributions that were not entirely modeled out earlier and traces of isopropanol (IPA) solvent, which ironically was used in an attempt to clean impurities such as the observed protein one. Note The fact that three components were extracted in all examples showed herein is not indicative of the range possible with SIMPLISMA. Some literature examples extract and analyze up to 1 components and more. [x 10 - ] 7 Infrared TMA-CO 3 TMA TMA Absorbance 5 3 1 IPA + hexane Protein contamination from H O bottle IPA + hexane IPA cleaning solvent 0 000 3800 300 300 300 3000 800 00 00 00 000 1800 100 100 100 1000 800 Wavenumber (cm-1) Figure 3: Resolved components in the aqueous photoresist developer.
It is somewhat ironic and interesting that water and isopropanol solvents, which are used in order to prevent impurities from appearing in the sample, are themselves sources of impurities. It suggests, for example, that one could perhaps allow more time to allow the IPA solvent to evaporate before loading a subsequent developer sample. SIMPLISMA was able to confirm the historic presence of impurities in the series, which could not have been done through study of each individual spectrum. Advantages The SIMPLISMA algorithm is particularly useful for mixture deformulation where the chemistry or spectroscopy of a sample is unknown (or only partially known) as is often the case for impurities. That is because it does not require a priori knowledge of the sample system. You don t need standards and you don t need reference materials to run SIMPLISMA. Another very important advantage of SIMPLISMA is that it can be used by anyone with the ability to interpret spectra. SIMPLISMA produces spectroscopic plots and results that can be responsibly and interactively evaluated by using spectroscopic judgment (without requiring deep chemometric or mathematical expertise). The interactive nature is practical when troubleshooting in industrial environments since it is difficult in such settings to ensure controlled experiments, obtain replicates, or avoid uncontaminated samples. It is possible to direct the procedure by using chemical knowledge of expected components, helping ensure good interpretation. The instant feedback provided by the spectroscopic plots extracted for each component means that the scientist can pro-actively recognize when things go wrong as the results unfold. This means that corrective actions can be taken immediately and makes it less likely that artifacts will be misinterpreted as real features. Since SIMPLISMA is based on multivariate principles, it can be used as an educative tool to introduce people to multivariate techniques. Furthermore, the results obtained from SIMPLISMA can provide extremely valuable guidance in preparing and interpreting more complex chemometric models. 5
Discussion and Summary Most material manufacturing processes involves mixtures, whether these are part of the actual product or related support materials. When impurities cause extra processing steps to be required or entire lots fail, halting the manufacturing process, quick troubleshooting and resolution of these issues can become business critical. In this context, scientists are often called to use instrumental analysis to determine the cause of the failure rapidly, because an idling manufacturing plant can represent loses in the order of millions of dollars. Whereas molecular spectroscopy often has sufficient sensitivity to detect small amounts of components, selectivity can be an issue particularly when analyzing mixtures. We showed the analysis of a photoresist developer where SIMPLISMA was able to detect and resolve components in mixtures for extremely challenging ATR-IR spectra. The results allowed the variance belonging to impurities to be characterized and identified, which is the first step towards controlling or eliminating them. As such, this makes SIMPLISMA a great tool for reproducibility and repeatability studies for finding batch-to-batch raw material variations and product contaminations, and permitting comparative analysis of a "good vs. bad" product batches. In a sense, SIMPLISMA bridges the gap between process and analytical chemists. On the one hand, the process people generally have a good understanding of expected sample composition, without necessarily understanding the underlying spectroscopy. On the other hand, spectroscopists can properly interpret spectral changes but don t always have insight on how process may affect the chemistry. It is to be noted that SIMPLISMA provided insight on reaction or aging kinetics in a fairly easy manner compared to more classical methods. References 1. Windig, W.; Guilment, J. Anal. Chem. 3, 15-13 (1991).. ACD/UV-IR Manager, version 10, Advanced Chemistry Development, Inc. Toronto, Canada, www.acdlabs.com/uvir/, 007. 3. Windig, W.; Stephenson, D.A., Anal. Chem., 735-7 (199).. Windig, W. Chemom. and Intell. Lab. Syst. 3, 3-1 (1997). 5. Bogomolov, A.; Hachey, M., Williams, A., Software For Interactive Curve Resolution Using SIMPLISMA, in Progress in Chemometric Research (Editor Alexey L. Pomerantsev), Nova Science Publishers, New York, Chapter 10,199-135 (005).