A new type of wireless neural recording front-end with PWM modulation and VCO-based ADC is proposed. The target signal is first encoded into binary levels with varying pulse width by PWM modulation, which intrinsically has much higher dynamic range compared with conventional voltage-mode instrumentation amplifiers. The front-end is thus not saturated by stimulation artifacts, enabling simultaneous recording and stimulation. A VCO-based ADC then digitizes the front-end output. The nonlinearity issue in VCO quantizers is resolved by the binary output since only two operation points are used. The inherent sinc anti-aliasing filter and first-order noise shaping are exploited to reject supply ripples and enhance resolution. The proposed architecture is digitally intensive, hence can achieve low-power and low-voltage operation in scaled technology. The theoretical analysis and simulation based on behavior models are presented in this paper.