This paper proposed a high efficient 5-class sleep staging XGBoost based algorithm using a single channel electroencephalogram (EEG). The sum of spectrum absolute value, instead of Fast Fourier Transform (FFT) based spectral energy, was proposed for feature extraction, which greatly reduced the complexity. The system is synthesized using 0.18 μ m, and experimental verification using Field-Programmable Gate Array (FPGA) based on Sleep-EDF database. An overall accuracy (ACC) of 79.1% is achieved, which a highest ACC of 86.3% has been measured with specified patient.