Point cloud registration has gained prominence in applications such as intelligent transportation and robot route planning. However, establishing complicated registration networks on resource-constrained devices remains a challenge. To address this issue, an efficient dynamic learning framework based on fine-grained knowledge distillation named SPDD (Single-Point & Dimension Distillation) is proposed. The algorithm leverages the teacher-student model paradigm and a dual-module distillation framework, guaranteeing that the student model has a weight distribution similar to the teacher model to retain registration correctness. The experimental results show that SPDD ensures registration accuracy while maintaining low computational complexity, making it easy to deploy on mobile devices.