Superdiffusion is a very interesting physical phenomenon existing widely in the real world. We propose a sufficient condition for the occurrence of superdiffusion on duplex networks by applying the generating functionology from combinatorial mathematics. Our result indicates that, with fixed values of intra-layer diffusion constants, superdiffusion emerges from a certain combination of different network topologies, which provides a topological mechanism to regulate the whole diffusion process. On a more general ground, we provide two linking strategies to expand the applicability of the above topological mechanism on real-world networks, and construct a class of specific duplex networks on which superdiffusion occurs. We further verify that diffusion with negative inter-layer correlation is faster than those with positive correlation from a typical duplex structure. Given the ubiquity of diffusion in the real world, our results constitute a significant step toward a better understanding of superdiffusion processes on multiplex networks.