L-Sort: An Efficient Hardware for Real-time Multi-channel Spike Sorting with Localization

TL;DR

L-Sort is a novel FPGA-based hardware system that enables real-time multi-channel spike sorting with localization, offering improved efficiency and accuracy for neural signal analysis in brain-computer interfaces.

Contribution

This paper introduces L-Sort, a hardware design that performs real-time multi-channel spike sorting using localization techniques, without storing raw signals, and demonstrates hardware efficiency improvements.

Findings

L-Sort achieves high detection and clustering accuracy on Neuropixel datasets.

L-Sort consumes less FPGA resources compared to existing spike sorting hardware.

L-Sort operates in real-time suitable for closed-loop BCI applications.

Abstract

Spike sorting is essential for extracting neuronal information from neural signals and understanding brain function. With the advent of high-density microelectrode arrays (HDMEAs), the challenges and opportunities in multi-channel spike sorting have intensified. Real-time spike sorting is particularly crucial for closed-loop brain computer interface (BCI) applications, demanding efficient hardware implementations. This paper introduces L-Sort, an hardware design for real-time multi-channel spike sorting. Leveraging spike localization techniques, L-Sort achieves efficient spike detection and clustering without the need to store raw signals during detection. By incorporating median thresholding and geometric features, L-Sort demonstrates promising results in terms of accuracy and hardware efficiency. We assessed the detection and clustering accuracy of our design with publicly available datasets recorded using high-density neural probes (Neuropixel). We implemented our design on an FPGA and compared the results with state of the art. Results show that our designs consume less hardware resource comparing with other FPGA-based spike sorting hardware.