Embodiments of 3D memory devices and methods for forming the same are disclosed. In an example, a method for forming a 3D memory device is disclosed. A peripheral circuit is formed on a first substrate. A first semiconductor layer is formed on a second substrate. A supporting structure and a second semiconductor layer coplanar with the supporting structure are formed on the first semiconductor layer. A memory stack is formed above the supporting structure and the second semiconductor layer. The memory stack has a staircase region overlapping the supporting structure. A channel structure extending vertically through the memory stack and the second semiconductor layer into the first semiconductor layer is formed. The first substrate and the second substrate are bonded in a face-to-face manner.