The closed-loop neural interface is an actively researched field with broad applications. However, it has a serious issue in that stimulation itself makes huge artifacts (~mV) in the recording system saturating the amplifier, contaminating bio-signal (~ 50𝛍𝛍V) and the post-analysis during a disturbed period. Several stimulation artifact removal (SAR) techniques were presented but have trade-offs between conversion time and algorithm accuracy. Here, we suggested a closed-loop neural interface with a fast convergence SAR algorithm while sustaining adequate removal accuracy. The amplifier-free ADC-direct 2nd-order continuous delta-sigma modulator recording stage is adopted to provide ~15𝛍𝛍 Vrms input-referred integrated noise from 5Hz to 5kHz which is adequate to record bio-signal. Also, it succeeds in recording ~10mV signals about the size of the stimulation artifact. The fast-convergence SAR module provides stable accuracy under the versatile recording environment. The chip is designed using the TSMC 65-nm CMOS process. The chip is composed of 18 input channels, 2 SAR blocks, one stimulation channel, and 2 CIC filters for down-sampling. The entire chip area is 1mm2.