A 32-Channel 85.4dB SNDR Time-multiplexed Neural Recording Frontend Achieving Within-Conversion Artifact Recovery

Next-generation large-scale implantable neuromodulation systems require a highdensity and ultra-low power recording front-end (RFE). Time-division multiplexing (TDM), sharing multiple channels with a single neural RFE, has proven effective in reducing per-channel area and power [1]–[5]. Their recent use in active microelectrode arrays has reduced aggregate wire count and packaging complexity [3]. TDM-based RFEs utilizing continuous-time incremental delta-sigma modulation (CT-IDSM) [3]–[5] for direct digitization effectively optimize the signal path, reducing silicon area further while delivering high-resolution data. These RFEs capture low amplitude neural signals (up to a few mVs) from channels, employing a separate servo loop to withstand electrode DC offsets (EDOs) up to tens of mVs. However, high-frequency stimulation artifacts (> 1kHz) and low-frequency motion artifacts (< 10Hz) limit the resolution of Nyquist samples in these systems. This degradation in signal quality is especially pronounced in large-scale closed-loop neuromodulation systems which experience frequent stimulation events.