For full functionality of this site it is necessary to enable JavaScript. Here are the instructions how to enable JavaScript in your web browser.
SIMS21, Poland 2017 - Spencer Angus Thomas abstract

Spencer Angus Thomas oral presentation (FN2-Tue1-1-5)

Towards autonomous analysis of 3D OrbiSIMS data

Spencer Angus Thomas, Alan M. Race, Ian S. Gilmore

National Physical Laboratory, National Physical Laboratory, Hampton Road, TW11 0LW Teddington, United Kingdom

The use of multivariate analysis have enhanced the mining of SIMS data, including; segmentation of images, dimensionality reduction, detection of regions of interest, and peak identification. Post-acquisition analysis can provide a wealth of information about the sample, however, may limit the extent to which a sample can be further explored. The volume of data that can be acquired in a single day means that the retrospective analysis can take a significant amount of time which creates a time lag for further investigations, for example a defect. Furthermore, sample and instrument availability may prohibit further analysis of identified regions of interest, such as a collection of peaks for MS/MS, or a subset of pixels in a SIMS image distinguishing chemical composition. Therefore the ability to mine these data during acquisition is essential to maximise the analysis information of a sample in a single experiment and improve workflow efficiency.

Using a defined pre-processing workflow and deep learning algorithms we can mine SIMS imaging data in a pixel-wise manor, i.e. individual spectra, which could be integrated into an online work during data acquisition. We demonstrate how this pipeline can enhance SIMS analysis by enabling automated region of interest detection or operational changes, such as locations for higher resolution images, MS/MS acquisitions, defect detection, or database matching. Specifically we demonstrate this with data from the new NPL 3D OrbiSIMS instrument and how this could be used for autonomous control enabling automated mode switching during a single experiment. The use of pre-trained deep neural networks analyse the data based on individual spectra providing fast and efficient analysis SIMS images. This occurs in nearer-real time and can analyse data during the acquisition. Moreover, this method is not limited by the number of pixels or the size of the data and therefore can process very large SIMS images where current methods fail.