The invention relates to the technical field of computer-aided design of integrated circuits, and provides a two-step X-architecture Steiner minimum tree construction method for very large scale integration (VLSI). Based on the advantages of an X-architecture model and a particle swarm optimization technique, the method is implemented through two steps: (1) the stage of social learning discrete particle swarm search, which comprises: using an edge-vertex encoding strategy capable of maintaining optimal topological information of particles, designing a fitness function taking wirelength into consideration; and using a chaotic decreasing mutation strategy and a new social learning strategy to design a new discrete particle swarm update formula; and (2) a stage of wirelength optimization, which comprises: designing a local topological optimization strategy to minimize the wirelength of an X-architecture Steiner tree. The method guarantees short total wirelength of nets and has high stability, thus being able to construct a high-quality X-architecture Steiner minimum tree.