A Real-Time Spike Sorting System Using Parallel OSort Clustering.

Area of Science:

• Neuroscience
• Computer Engineering
• Signal Processing

Background:

• Real-time spike sorting is crucial for understanding neural activity.
• Existing methods often face limitations in latency and computational efficiency for in-vivo applications.

Purpose of the Study:

• To design and implement an efficient, real-time spike sorting system using unsupervised clustering.
• To adapt the online sorting (OSort) algorithm for hardware realization with reduced latency and memory requirements.

Main Methods:

• Utilized the online sorting (OSort) algorithm, modeled in both floating-point and fixed-point representations.
• Proposed a modified parallel OSort algorithm to reduce memory accesses and computations.
• Implemented a novel memory configuration scheme for parallel processing on a Xilinx Artix-7 FPGA.

Main Results:

• Achieved significantly reduced classification/clustering latency, enabling in-vivo spike sorting.
• Demonstrated an efficient hardware realization of the OSort algorithm.
• Estimated ASIC implementation: 2.57 mm², 2.78 mW power consumption at 24 kHz.

Conclusions:

• The developed OSort-based spike sorting system offers an efficient and low-latency solution for real-time neural data processing.
• The hardware architecture is suitable for in-vivo applications requiring high-throughput spike detection and classification.
• The system's performance on FPGA and projected ASIC metrics highlight its practical viability.