Photoelectrochemical multi-junction devices for direct solar energy conversion have been highly improved during the last decade with solar to hydrogen efficiencies reaching almost up to 20%. However, these efficiencies are still below their expected physical limits, which requires a deeper understanding of band energy diagrams along the functional device interfaces in the vicinity of a liquid electrolyte in order to identify potential and charge transfer losses, that will limit the conversion efficiency of the overall device. For this purpose, model surfaces of classical elemental (Si) and binary (InP) semiconductors were prepared and characterized by photoemission spectroscopy with respect to their electronic structure and electronic surface state formation. The interaction of these surfaces with water was investigated by modeling the electrochemical interface in ultra-high vacuum using a “frozen electrolyte” approach. Depending on surface termination and surface state concentration, the surfaces showed a shift in Fermi level towards the vacuum level, indicating an electron injection upon water adsorption by the interaction with unsaturated dangling surface bonds. The contact formation of the photoabsorber to the noble-metal catalyst results in an electron depletion layer acting as a barrier for the charge transfer and therefore preventing considerable conversion efficiencies. Using TiO₂ as buffer layer in between the photoabsorber and catalyst seemed to prevent the strong depletion of the photoabsorber. However, this effect strongly depends on the TiO₂ film properties, which results from the preparation process. This has to be optimized in order to guarantee a loss-free charge transfer from the photoabsorber to the catalyst. The deduced energy band diagrams from modeled interface experiments help to understand the electrochemical performance when using the layer arrangement in a device-like setup. However, when method-related microstructural effects like lateral inhomogenities or mechanical and structural stability come into play, the performance prediction solely derived from the energy band diagrams of model interfaces seem to fail and cannot fully describe the energetic device complexity at operation conditions.