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

Promovendus/a: Davit Melkonyan

Promotor(en): Prof. dr. ir. Wilfried Vandervorst

The technological advancements of the past several decades have revolutionized almost every aspect of our life. The rapid growth of the semiconductor industry was one of the main enablers of these advancements as it allowed to fabricate integrated circuits which paved the way for the emergence of computers and smartphones. This growth was achieved by scaling of the semiconductor device dimensions, comprising the integrated circuits, which increased their performance decreasing the price. However, the drastic scaling of the device dimensions and the implementation of new device architectures and materials requires a detailed structural understanding of the atomic composition, imposing significant challenges on metrology. Ideally, one should be able to assess the atomic composition in 3D with an atomic spatial resolution, high sensitivity towards low dopant concentrations in a fully quantitative way. In this respect, the emerging technology of atom probe tomography (APT) has the potential to become well established in the field of advanced semiconductor technology due to its 3-D nature and high chemical sensitivity. In APT, atoms and molecules constituting the specimen are field evaporated as ions, from the tip of a needle-like specimen and subsequently collected on a position-sensitive single particle detector. Combining the position-sensitive nature of the detection system and the ion arrival sequence, a three-dimensional map of the real-space specimen ion positions are reconstructed in 3-D visualization software. In this thesis we explored the possibility of APT analysis of embedded semiconductor devices, such as FinFETs, to map their constituent atoms in 3-D with atomic resolution. It is challenging, as in many cases APT suffers from biased composition and erroneous dimensions which are more pronounced for heterogenous systems such as semiconductor devices. In addition, APT requires nanoscale needle-shaped specimen, containing the region of interest, the fabrication of which is also a challenge. We used B doped Si samples to study some of the fundamental causes of compositional biases. We found that depending on the experimental parameters the measured B concentration can be either underestimated or overestimated. This originates from the electric field differences required to evaporate B and Si atoms. B atoms require more electric field to evaporate compared to Si which leads to the B retention at the specimen surface leading to uncontrolled B evaporation. This, in turn, induces detector pile-up inducing preferential B underestimation. The retention also causes B migration on the evaporating surface leading to B overestimation and artificial long profile tails on steep B profiles. The dependence of these phenomena on experimental parameters was investigated and means to rectify or mitigate the observed artefacts are offered. Besides the quantification errors, one suffers from severe erroneous dimensions or unrealistic density variations in the reconstructed volume. We used SiGe fins embedded in SiO2 to study these geometrical artefacts. Here also the evaporation field variations between the SiGe and SiO2 results in preferential evaporation of SiGe, which develops lower local radii throughout the evaporation. This, in turn, affects the trajectories of evaporated ions causing lower magnification for fin compared to SiO2. As a result, the reconstructed fin is much smaller than the original one. Using complementary TEM measurements we developed an algorithm to correct this dimensional distortion.

Finally, we addressed some practical challenges concerning specimen preparation. We incorporated a wet chemical etching to the standard focused ion beam (FIB) milling sample preparation procedure. This allowed to easier localize nanowires and fins at a specimen apex. The method also can be used to preferentially remove materials from the specimen prior to the analysis that are difficult to analyze such as SiO2, improving the data quality of the remaining components.