A nonparametric neural signal processor for online data compression and power management

This paper reports a 8-channel neural spike processor to permit unsupervised signal processing, substantial bandwidth reduction, and automatic power management in extracellular neural recording experiments. In this work, spikes are detected based on their proportions in real-Time estimated power density function of neural data, which provides a reliable prediction of spiking activities measured in probabilities. A closed-loop control has been designed by estimating firing rates based on alignment results and used to selectively turn on recording channels and signal processing modules. The proposed system was implemented in a 0.13 μm CMOS technology and has a varied power dissipation from 36 μW to 54.4 μW per channel at a voltage supply of 1.2 V. The chip can be configured in various output modes to meet different application needs and provides a over 180× data rate reduction. The system functionalities and performances have been verified by both benchtop testing and in vivo animal experiment.