A new CORDIC algorithm based on sign bits prediction is presented. It is an effective method for transforming arbitrary iterations of classic CORDIC to simple matrix multiplications. At the same time, this algorithm is implemented based on synchronized data triggering architecture (SDTA). Experimental results show that the average speed-up of transcendental functions tested based on our version CORDIC compared with mathematic library supported in gcc3.2.2 is up to 283% under the given Neuron microprocessor.