Strategies For Trace Metal Quantification In Polymer Samples With An Unknown Matrix Using Laser-Induced Breakdown Spectroscopy

Providing unique advantages, laser-based analytical techniques such as LIBS have gained more and more popularity for quantitative elemental analysis in the last few years. However, to obtain reliable quantitative results, matrix-matched standards are required. A particular material of interest for quantitative trace metal analysis is synthetic polymers, which is among the most widely used materials in our modern world. As the exact composition of a polymer under investigation (polymer type and applied additives) is often not known, the selection of an appropriate matrix-matched standard is difficult. In this work, we investigate and assess different approaches for quantifying potassium in unknown polymer types or polymers with an unknown composition where matrix-matched standards cannot be employed. This is of great interest in the semiconductor industry where monitoring of mobile ions in applied polymers is crucial, and the composition of the polymer is often not known due to confidentiality. We use the unique capabilities of LIBS, providing adequate sensitivity for potassium, and additionally delivering polymer-specific emission signals. Two different multivariate approaches (Random Decision Forest classification combined with conventional univariate calibration and a Partial Least Squares model) are developed and applied. Therefore, an in-house prepared library of standards of 8 different polymer types (Acrylic, PAN, PI, PMMA, PSU, PVA, PVC and PVP) is prepared. The errors obtained from the multivariate approaches are compared with conventional matrix-matched as well as non-matrix-matched quantification. With our developed approaches, for some samples quantitative determination of potassium in the low mu g/g range in unknown polymer types is achieved with a relative error less than 20% which is comparable to conventional matrix-matched quantification. For all other samples, relative errors in the range of 30%-90% are obtained, which offers a precision adequate for many applications. E.g. in the food-packaging- or semiconductor-industry in many cases it is sufficient to determine if the contamination level of a sample with an unknown matrix is below or above a certain threshold. In this case, the developed approach poses a significant improvement compared to non-matrix-matched quantification which often leads to deviations up to a factor 10 or more from the nominal concentration.