Analog-to-digital conversion (neural ADC)

The analog-to-digital converter turns each amplified, filtered channel into numbers. Neural recording sets its specs: enough resolution (commonly 10 to 16 bits) to span the dynamic range from small LFP to large spikes and artifacts, a sample rate of at least about 20 to 30 kilosamples per second per channel to capture spike waveforms (much lower if only LFP is kept), and power low enough to fit the implant budget when multiplied across all channels.

Architecture choice trades those axes. Successive-approximation (SAR) converters are compact and energy-efficient per conversion, which makes them popular as per-channel or shared-column ADCs. Delta-sigma converters oversample to reach high resolution and are well suited to lower-bandwidth LFP-grade recording. Neuropixels, for instance, uses column-parallel ADCs shared across many electrodes.

One ADC is frequently time-shared across many channels through a multiplexer, so the converter must settle to full accuracy within each channel's brief time slot. Designers compare converters with energy-per-conversion figures of merit (Walden and Schreier), because at high channel counts the ADC's share of total power — and hence of the thermal budget — is not negligible.