Two-dimensional (2D) transition metal dichalcogenides (TMDs) present varied optical and electronic properties associated with their different crystal structures. We optimize here the chemical vapor deposition (CVD) of tungsten-based TMDs, such as hexagonal (1H) WSe2 and WS2 monolayers, as well as monoclinic (1T’) WTe2. While 1H WSe2 and WS2 are typically direct band-gap semiconductors in the monolayer form, the anisotropic 1T' WTe2 is a small band gap topological insulator. Our purpose is to study the structural and electronic phase transition of ternary alloy WSe2xTe2(1-x) between the 1H and 1T' crystal structures. However, the synthesis of the mid-composition (x~0.5) alloys by CVD remains challenging. By just adding an additional monolayer, we obtain bilayer (2ML) structures, in which the available polytypes combinations, and the relative physical properties, are even richer. We further optimize our CVD growth methods to maximize the yield of 2ML hexagonal WSe2 and WS2. In addition to Raman and photoluminescence spectroscopies, we employ various other experimental techniques, including second harmonic generation and selective area electron diffraction, to discern between the 2ML polytypes. We then focus on the 3R bilayer stacking of WSe2 and we resolve its electronic band structure by angle-resolved photoemission spectroscopy.

Abstract 

This thesis focuses on the modeling of narrow-gap far-infrared semiconductors, ternary alloys, and heterovalent heterostructures. To overcome the limitations of conventional Density Functional Theory (DFT), a Bayesian Machine (BMach) framework [2] is developed to optimize internal DFT meta-parameters, such as the Hubbard U in DFT+U and the mixing (α) and screening (μ) parameters in hybrid exchange–correlation functionals. These parameters are benchmarked against quasiparticle corrections obtained from Green’s function–based quasiparticle-GW (G₀W₀) calculations, with BMach employing Gaussian process regression (GPR) and physics-informed loss functions.

Applied to the benchmark semiconductors InSb and InAs, BMach accurately reproduces experimental band gaps, effective masses, and Luttinger parameters at significantly reduced computational cost [1]. The study further explores CuPt-ordered InAsₓSb₁₋ₓ alloys across the full composition range using both supercell and Virtual Crystal Approximation (VCA) approaches; notably, the VCA-based G₀W₀ method yields a bowing parameter of 0.838 eV, in excellent agreement with experiment. Finally, the band alignment at the CdTe/InSb(001) heterointerface is determined using the potential-lineup method with BMach-calibrated parameters, confirming a type-I alignment consistent with experimental observations.

This work establishes a transferable and extensible computational framework for high-fidelity electronic-structure predictions across diverse material systems, with direct relevance for future electronic, optoelectronic, energy, and quantum technologies.

Abstract 

In this PhD thesis, we advance integrated photonics for quantum information processing by improving the calibration and control of photonic integrated circuits (PICs). PICs enable compact and stable light manipulation for a wide range of applications in optics. We demonstrate quantum protocols on specialized PICs, including the first on-chip certified randomness generation and high-fidelity 4-GHZ state tomography. In universal PIC architectures, we develop a machine learning-assisted characterization technique to mitigate hardware imperfections, achieving a record 99.77% fidelity in unitary operations on a 12-mode interferometer. Finally, we refine crosstalk models and propose a robustness criterion for interferometer design, enhancing PIC control accuracy. Our results, including a patented machine learning-based method, contribute to both quantum and classical integrated photonics, advancing scalable photonic quantum computing.

Figure : Photonic integrated circuits (PICs) consist of waveguides (blue lines) guiding light across a slab of transparent material. PICs often exhibit various imperfections resulting from fabrication constraints, tolerances, and operation wavelength, illustrated here on a simplified PIC. In general, input and output ports have different optical transmissions. In addition, the actual beamsplitter reflectivity values deviate from the target. Phase shifters (purple components) dissipating heat entail that all the implemented physical phase shifts depend on all the applied voltages. Finally, optical path variations lead to non-zero phase shifts even without any voltages applied. We use machine-learning to find suitable values for each of these parameters.