This paper presents a Spiking Diffusion Policy (SDP) framework for robotic manipulation. The SDP framework integrates Spiking Neurons and Learnable Channel-wise Membrane Thresholds (LCMT) into the diffusion policy model, significantly enhancing computational efficiency while maintaining high performance. Specifically, the proposed SDP model employs a U-Net architecture constructed using a Spiking Neural Network (SNN) as the foundational network for the diffusion process. In the Spiking Diffusion Policy model, activations are transmitted in the spike form, substantially reducing the number of floating-point operations required by the network. To further improve the performance of Spiking Neural Networks in robotic manipulation tasks, we introduce LCMT to the Spiking Diffusion Policy, which aligns membrane potentials and spike firing rates across different channels, thereby mitigating the challenges and performance degradation associated with manually tuning such fine-grained hyperparameters. The proposed SDP model is evaluated on both simulated tasks and real-world scenarios, demonstrating comparable results with traditional artificial neural network counterparts. Additionally, our approach is projected to reduce dynamic energy consumption by approximately 80.36% on 45 nm hardware.