A Multi-Channel Spike Sorting Processor With Accurate Clustering Algorithm Using Convolutional Autoencoder.

Area of Science:

• Neuroscience
• Computer Engineering
• Signal Processing

Background:

• Accurate spike sorting is crucial for analyzing neural data.
• Existing methods face challenges in accuracy and computational efficiency.

Purpose of the Study:

• To develop an efficient spike sorting processor with high clustering accuracy.
• To implement a novel spike sorting algorithm in hardware.

Main Methods:

• Utilized an L2-normalized convolutional autoencoder for feature extraction.
• Employed a similarity-based K-means clustering algorithm with cosine similarity.
• Designed an efficient time-multiplexed hardware architecture in 40-nm CMOS.

Main Results:

• Achieved 95.54% clustering accuracy, outperforming previous designs.
• Processor power consumption is 224.75μW/mm² at 7.68 MHz and 0.55 V for 16 channels.
• Demonstrated improved convergence and online clustering capabilities.

Conclusions:

• The proposed spike sorting processor offers a significant advancement in neural signal processing hardware.
• The novel algorithm and hardware co-design enable high-accuracy, low-power neural data analysis.