Two methods are used to compute the STAP weights in application, named as data domain algorithm and mean square domain algorithm, which are represented by QR decomposition algorithm and sample matrix inversion (SMI) algorithm respectively. In comparison, SMI algorithm can be realized simply with sample covariance matrix inversion, QR decomposition algorithm can be mapped onto systolic array with a high design cost. The inner parallelism of different algorithms of matrix inversion is reviewed firstly, the implementation of SMI method based on single DSP is proposed based on the on-chip parallel technique, the numerical stability of the proposed method is also given, and experiments with the data measured verify the validity of the proposed method.