Hardware architecture for on-chip unsupervised online neural spike sorting

Microelectrode arrays can acquire neural signals in parallel from multiple channels. Spike sorting has emerged as one of the most significant challenges in multichannel systems. An ideal spike sorting system must be implantable, unsupervised, online and scalable to hundreds of channels. This paper proposes a novel hardware architecture for on-chip and unsupervised neural spike sorting with Teager Energy Operator detection, Zero-Crossing Features and an online clustering algorithm, MCK Classifier, which is a modification of the standard K-Means. The reported classifier gives an average detection-classification accuracy of 82% at typical SNR of 7dB, which is within 2% of the standard K-Means classifier.