The　obstacle-avoiding　rectilinear　Steiner　minimum　tree　(OARSMT)　problem　is　one　of　the　fundamental　problems　in　the　physical　design　of　Very　Large　Scale　Integrated　(VLSI)　circuits,　especially　in　routing,　which　is　known　to　be　NP-complete.　It　becomes　more　important　than　ever　for　VLSI　designs　which　need　to　consider　numerous　obstacles　such　as　macro　cells,　IP　blocks　and　pre-routed　nets.　Particle　Swarm　Optimization　(PSO)　has　been　proved　to　be　an　efficient　intelligent　algorithm　for　optimization　designs.　In　this　paper,　an　improved　Discrete　Particle　Swarm　Optimization　(DPSO)　algorithm　is　proposed　for　solving　the　OARSMT　problem　optimally.　Meanwhile,　the　corresponding　evaluation　function　and　the　operators　inspired　from　the　Genetic　Algorithm　(GA)　are　designed.　The　obstacle-avoiding　measure　is　then　presented　to　construct　an　OARSMT,　which　avoiding　obstacles　by　generate　virtual　vertexes　of　the　real　vertexes.　Experimental　results　on　several　benchmarks　show　that　the　proposed　method　achieves　good　performance　in　routing　optimization　significantly.　©　2011　IEEE.