FPGA-based clustering of multi-channel neural spike trains

Electro-physiological recording of neural bioelectrical activity contains local field potentials and unit activities. Unit activity is a mixture of action potentials generated by the neurons. Spike sorting is a method to determine which individual neurons produce the recorded unit activity. High-channel-count neural probes can measure more than a hundred different positions of the brain in parallel, so large amount of highdimensional data is generated. To increase the computational speed and decrease the processing time Field-Programmable Gate Array (FPGA) architectures can be applied as hardware accelerators. In this paper an FPGA-based implementation of the Expectation-Maximization (EM) algorithm for neural spike clustering is presented.