An ultra low-power 16-channel wireless neural recording system-on-a-chip (SoC) is implemented in 0.18μm standard CMOS process. A novel topology of analog front-end (AFE), which combines the auto-zero technique and channel switching method to obtain both low-noise and low-power features, is presented in this work. A time-domain-based modulation approach reduces the power consumption of wireless transmitter dramatically by 80%. A RF energy harvester is also employed in this chip for sustainable use. The input referred noise and average power per channel of AFE are 0.5 μV_rms and 1.345 μW respectively. The whole system consumes only 473 μW while performing the signal recording and wireless transmission. Furthermore, practical experiments of human EEG signal recording are demonstrated to verify the functions of proposed SoC.