Constructing　a　timing-driven　Steiner　tree　is　very　important　in　VLSI　performance-driven　routing　stage.　Meanwhile,　non-Manhattan　architecture　is　supported　by　several　manufacturing　technologies　and　now　well　appreciated　in　the　chip　manufacturing　circle.　However,　limited　progress　has　been　reported　on　the　non-Manhattan　performance-driven　routing　problem.　In　this　paper,　an　efficient　algorithm,　namely,　TOST_BR_MOPSO,　is　presented　to　construct　the　minimum-cost　spanning　tree　with　a　minimum　radius　for　performance-driven　routing　in　Octilinear　architecture　(one　type　of　the　non-Manhattan　architecture)　based　on　multi-objective　particle　swarm　optimization　(MOPSO)　and　Elmore　delay　model.　Edge　transformation　is　employed　in　our　algorithm　to　make　the　particles　have　the　ability　to　achieve　the　optimal　solution　while　Union-Find　partition　is　used　to　prevent　the　generation　of　invalid　solution.　For　the　purpose　of　reducing　the　number　of　bends　which　is　one　of　the　key　factors　of　chip　manufacturability,　we　also　present　an　edge-vertex　encoding　strategy　combined　with　edge　transformation.　To　our　best　knowledge,　no　approach　has　been　proposed　to　optimize　the　number　of　bends　in　the　process　of　constructing　the　non-Manhattan　timing-driven　Steiner　tree.　Moreover,　the　theorem　of　Markov　chain　is　used　to　prove　the　global　convergence　of　our　proposed　algorithm.　Experimental　results　indicate　that　the　proposed　MOPSO　is　worthy　of　being　studied　in　the　field　of　multi-objective　optimization　problems,　and　our　algorithm　has　a　better　tradeoff　between　the　wire　length　and　radius　of　the　routing　tree　and　has　achieved　a　better　delay　value.　Meanwhile,　combining　edge　transformation　with　the　encoding　strategy,　the　proposed　algorithm　can　significantly　reduce　nearly　20 %　in　the　number　of　bends.　©　2014,　Springer-Verlag　Berlin　Heidelberg.