An Unsupervised and Real-Time Spike Sorting Processor based on Digital Perturbed K-means Spike Sorting

Next-generation spike sorting processor needs to deal with extremely high data input. Hence a high-accuracy and low-area processor is critical for multi-channel implantable neural recording systems. In this study, we propose a spike sorting processor based on a novel architecture that runs unsupervised and in real-time. It features an adaptive-threshold spike detection that is based on a Sigma-Delta Control Loop and a novel spike classifier based on perturbed K-means. The test results show that the spike sorting processor has achieved an accuracy of over 93.2% on widely used synthetic datasets and real neural signal tests. The system has been synthesized using a 180-nm CMOS process, and our resource analysis reveals that it consumes just $2.56 \mu \mathrm{~W}$ per channel with a $1.8-\mathrm{V}$ supply, with a core area of $0.056 \mathrm{~mm}^{2} \mathrm{per}$ channel.