Atomic scale observation of atom distributions in 3D devices using atom probe tomography

The rapid growth of the semiconductor industry over the past several decades was enabled by scaling of the integrated circuits to smaller physical dimensions, which allowed to increase their performance simultaneously decreasing their cost. However, due to this decrease of dimensions, the complexity of semiconductor devices had to be increased to preserve and improve their functionality. New device architectures and additional materials are continuously incorporated into semiconductor devices, which put stricter requirements on metrology. For optimal manufacturing process control of contemporary semiconductor devices, 3D metrology with close to atomic scale resolution and chemical sensitivity up to one ppm is required. Atom probe tomography is a developing technique that has the potential to meet these demands. In this thesis, we study and advance the ability of APT to analyze the state-of-the-art semiconductor devices by addressing the main challenges and their root causes. We discuss the compositional and structural analyses and cover some practical aspects such as specimen preparation, using relevant test structures and field evaporation simulations.

The compositional analysis is one of the main requirements as the concentration of impurities and alloy elements have a crucial impact on device performance. The main mechanisms leading to biased composition induced by APT are briefly reviewed, some of which are studied in detail. Firstly, we focused on the error source being the long-ranging transient originating from the evaporation field difference between high evaporation field dopants and low evaporation field matrix such as B doped Si. We showed that the root of this artefact lies in a dynamic process of atom migration, dome formation and smoothening occurring at the evaporating surface. This artefact manifests itself as serious spatial distortions and severe deviation from stoichiometric emission (and thus correct quantification) leading to long transient tails on steep B profiles. This transient is only restored by a dynamic enrichment of B at the evaporating surface, whereby the lower B evaporation probability (due to its high evaporation field threshold) is counterbalanced by an increase of potential evaporation sites (due to B-surface enrichment). The width of the transient region depends on the nominal B concentration of the sample and can reach 10s of nanometers. Optimal conditions to minimize the adverse impact of this artefact is determined. Secondly, the multi-hit defectivity (and means to correct them) is discussed. It was shown that the B atoms preferentially evaporate as multi-hit events, the combination of which with the limited capacity of APT to resolve the multi-hit events, known as ion pile-up, lead to B underestimation. The B loss due to multi-hit events was quantified, and a detailed analysis of the different isotopic mixes of the B-B double hit pairs was used to recover the underestimated B, which we demonstrate comparing APT and SIMS measurements.

Another important aspect of device characterization is the analysis of complete devices. In this aspect, APT suffers greatly from local magnification effects, which induce dimensional distortions and anomalous density variations in reconstructed data. Using a 40 nm wide SiGe fin embedded in SiO2 as a test structure, we discussed the root cause of these artefacts. Additionally, we propose a simple data treatment routine, relying on complementary transmission electron microscopy analysis, to improve compositional analysis of the embedded SiGe fins. Using field evaporation simulations, we show that for high oxide to fin width ratios, the difference in evaporation field thresholds between SiGe and SiO2 results in a non-hemispherical emitter shape with a negative curvature in the direction across, but not along the fin. This peculiar emitter shape leads to severe local variations in radius and hence in magnification across the emitter apex, causing ion trajectory aberrations and crossings. As shown by our experiments and simulations, this translates into unrealistic variations in the detected atom densities and faulty dimensions in the reconstructed volume, with the width of the fin being up to six-fold compressed. Rectification of the faulty dimensions and density variations in the SiGe fin was demonstrated with our dedicated data treatment routine.

 The specimen preparation is another important aspect for successful APT analysis and can majorly affect its yield. We introduce an innovative specimen preparation method employing the selectivity of a wet-chemical etching step to improve data quality and success rates in the atom probe analysis of contemporary semiconductor devices. Firstly, on the example of a SiGe fin embedded in SiO2, we demonstrate how the selective removal of SiO2 from the final APT specimen significantly improves the accuracy and reliability of the reconstructed data. With the oxide removal, we eliminate the origin of the above discussed local magnification effects. Secondly, using the same approach, we increase success rates to ~90% for the damage-free, 3D site-specific localization of short (250 nm), vertical Si nanowires at the specimen apex. The Ge content within a SiGe fin as well as the 3D B distribution in a Si NW as resolved by APT analysis are in good agreement with TEM/EDS and ToF-SIMS analysis, respectively.