Abstract: Carbon ion radiotherapy suffers from range uncertainty due to differences in human tissue density, patient organ motion, and positioning errors, so increasing the range margin is usually used to set up the planned or clinical target area during treatment planning, but this method increases the damage to the surrounding normal tissues, and in addition to this uncertainty, it may also result in potentially excessive range deviation, which also increases the damage to healthy tissues. In order to be able to reduce the range margin to lower the dose to non-tumour tissues as well as to provide feedback by interlocking signals via digital I/O when the range deviation is larger than a set threshold, a fast Bragg peak prediction scheme for FPGA-based CeBr₃ scintillator crystal arrays has been proposed by using a transplantation strategy of machine learning algorithms, and the algorithms can complete the depth prediction in 15.31 μs, using 46 ms of cumulative data from the detector system. The average error of the feedback accuracy of this scheme is tested to be only 0.011 mm.