As device architectures become increasingly complex and heterogeneous, advanced three-dimensional (3D) characterization tools are needed to elucidate structure-property relationships and optimize process conditions. At the nanoscale, high-angle annular dark-field scanning transmission electron microscope (STEM-HAADF) tomography is widely used for 3D morphological analysis of semiconductor devices [1]. This technique consists in acquiring a series of 2D STEM-HAADF projections at different viewing angles, and applying a dedicated algorithm to retrieve the 3D morphology of the object. 3D chemical imaging of nanodevices is nowadays possible by energy-dispersive x-ray spectroscopy STEM (STEM-EDX) tomography, which has benefited greatly from recent developments in electron sources such as the X-FEG (Field Emission Gun), and multiple X-ray detector systems such as the Super-X, incorporating four SSD (Silicon Drift Detectors) detectors [2]. The technique remains however very time-consuming, and reduced X-ray count rate and number of projections are necessary to minimize the total acquisition time and avoid beam damage during the experiment. In addition, tomographic series of STEM-EDX datacubes are too large to be analyzed by commercial software packages in an optimal way. In order to make STEM-EDX tomography accessible to the semiconductor industry, it is necessary to automate the data processing and employ sophisticated methods capable of producing high quality reconstructions from a limited number of noisy projections. In this presentation, we will give an overview of the recent advances in electron tomography, with an emphasis on STEM-EDX tomography and the processing tools necessary for accessing reliable 3D information. We will show that multivariate statistical analysis methods [3] can be used for unsupervised identification of chemical phases, while automated elemental analysis of very noisy EDX-STEM datasets can be achieved by principal component analysis (PCA) followed by Gaussian curve-fitting methods. We will also illustrate the superior performance of compressed sensing (CS) approaches [4,5] compared to classical tomographic algorithms, for 3D reconstructions from highly under-sampled datasets. As an example, we show below the 3D chemical analysis of an arsenic-doped silicon structure for fin field-effect transistor (FinFET) technology [6]. Other applications of electron tomography for the 3D analysis of semiconductor devices and materials will be presented, ranging from phase-change materials [5] to DNA origami nanostructures for silicon patterning. Prospects of deep learning approaches for spectral analysis and tomographic reconstruction will be also discussed.