In　this　paper,　an　optimization　method　for　ring　oscillator　(RO)-based　thermal　sensor　allocation　based　on　a　fuzzy　clustering　algorithm　is　proposed.　A　sensor　array　is　distributed　uniformly　on　a　field　programmable　gate　array　(FPGA)　as　an　original　layout.　Using　a　trade-off　between　sensor　numbers　and　measurement　resolution,　the　thermal　sensors　are　combined　based　on　their　mutual　clustering　levels,　which　are　inversely　proportional　to　the　temperature　difference.　This　method　is　suitable　for　various　applications　on　FPGAs.　A　self-heating　test　module　that　uses　the　FPGA　resources　to　generate　hotspots　and　large　temperature　gradients　for　measurement　validation　is　presented.　The　experimental　results　show　that　when　the　temperature　gradients　range　from　0　degrees　C　up　to　40　degrees　C　on　an　Altera　Cyclone　IV　FPGA,　we　can　reduce　sensor　resource　requirements　by　37.5%　while　maintaining　a　measurement　resolution　of　1　degrees　C.