A vertical stack of a first silicon germanium alloy layer, a second epitaxial silicon layer, a second silicon germanium layer, and a germanium layer are formed epitaxially on a top surface of a first epitaxial silicon layer. The second epitaxial silicon layer, the second silicon germanium layer, and the germanium layer are patterned and encapsulated by a dielectric cap portion, a dielectric spacer, and the first silicon germanium layer. The silicon germanium layer is removed between the first and second silicon layers to form a silicon germanium mesa structure that structurally support an overhanging structure comprising a stack of a silicon portion, a silicon germanium alloy portion, a germanium photodetector, and a dielectric cap portion. The germanium photodetector is suspended by the silicon germanium mesa structure and does not abut a silicon waveguide. Germanium diffusion into the silicon waveguide and defect density in the germanium detector are minimized.