Embodiments of methods for forming three-dimensional (3D) memory devices are disclosed. In an example, a peripheral device is formed on a first substrate. A first interconnect layer is formed above the peripheral device on the first substrate. A dielectric stack including a plurality of dielectric/sacrificial layer pairs and a plurality of memory strings each extending vertically through the dielectric stack is formed on a second substrate. A second interconnect layer is formed above the memory strings on the second substrate. The first substrate and the second substrate are bonded, so that the first interconnect layer is below and in contact with the second interconnect layer. The second substrate is thinned after the bonding. A memory stack is formed below the thinned second substrate and including a plurality of conductor/dielectric layer pairs by replacing, with a plurality of conductor layers, sacrificial layers in the dielectric/sacrificial layer pairs.