EDX‐STEM phase mapping of semiconductor devices using multivariate statistical analysis tools

Abstract

With the increasing miniaturization of electronic devices, high resolution structural and analytical characterization tools are necessary for the optimization of fabrication processes. Scanning transmission electron microscopy energy dispersive X‐ray (EDX‐STEM) spectroscopy is a well‐established technique that has recently gained momentum thanks to the introduction of high‐brightness electron sources and the Super‐X EDX system (4 SDD detectors), allowing fast EDX mapping with high collection efficiency. While the traditional EDX data analysis consists in extracting the elemental map of each element present in the sample [1], it is often the case that the aim of the analysis is to investigate the spatial distribution, shape and thickness of the different chemical phases present in the sample. Multivariate statistical analysis tools, such as non‐negative matrix factorization (NMF) and independent component analysis (ICA), were shown to yield simplified interpretation of spectral datasets by rapid identification of phases (e.g. [2,3]). In this work, we applied NMF to the EDX analysis of Si/SiGe multilayers to validate the Sidewall Image Transfer (SIT) process developed for their patterning [4]. A FIB‐prepared lamella was characterized in an FEI Titan Themis operating at 200kV and equipped with a probe corrector and 4 SDD EDX detectors. An EDX‐STEM map was acquired with TIA, using a pixel size of 1nm and a dwell time of 20ms/pixel, and exported to hyperspy, a python‐based software for hyperspectral data processing [5]. Spectral unmixing using NMF led to the identification of five chemical phases in the sample: Si, SiGe, SiO2, TiN and C (see the component spectra in Figure 1 and the corresponding loadings in Figure 2(a‐e)). More specifically, NMF succeeded in: (1) separating the Si signal emanating from pure Si, SiGe and SiO2 layers; and (2) deconvoluting the C, N and O peaks. This greatly simplified the compositional analysis (Figure 2(f)), and allowed a more straightforward estimation of the thickness of the different layers, as shown in Figure 2(g). NMF combined to EDX‐STEM tomography was recently applied to superalloy systems for aerospace applications [6,7]. We will show that this approach has also the potential to address materials characterization challenges currently facing the semiconductor industry, such as the chemical analysis of dopants and impurities [8].