A Custom dual-processor System for Real-time Neural Signal Processing

This paper presents a custom dual-processor SoC architecture, studied and customized to support information extraction from signals acquired from Peripheral Neural System, for prosthetic applications. The main tasks accomplished by the processing implemented on the computing platform are noise removal and identification of neural spikes. On-board execution of such tasks allows to identify which samples actually contain useful information. Thus, it reduces required input/output bandwidth, so that connection to the external environment can be implemented using a Bluetooth Low Energy device. The overall SoC architecture has power consumption compliant with implant-related constraints with a battery lifetime of around one-day.