Density-based hardware-oriented classification for spike sorting microsystems

Successful proof-of-concept laboratory experiments on cortically-controlled brain computer interface motivate continued development for neural prosthetic microsystems (NPMs). One of the research directions is to realize realtime spike sorting processors (SSPs) on the NPM. The SSP detects the spikes, extracts the features, and then performs the classification algorithm in realtime in order to differentiate the spikes for the different firing neurons. Several architectures have been designed for the spike detection and feature extraction. However, the classification hardware is missing. To complete the SSP, a density-based hardware-oriented classification algorithm is proposed for hardware implementation. The traditional classification algorithms require a considerable memory space to store all the training features during the processing iteration, which results in a considerable power and area for the hardware. The proposed one is designed based on the density map of the spike features. The density map can be accumulated on-line with the coming of the spike features. Therefore the algorithm can save significant memory space, and is good for efficient hardware implementation.