A programmable and implantable microsystem for multimodal processing of ensemble neural recordings

Conditioning raw neural signals recorded through microelectrode arrays implanted in the brain is an important first step before information extraction can take place. This paper reports on the design and implementation of a programmable and fully implantable microsystem that fulfills this purpose. The system design builds on our earlier work that relies on a sparse representation of the neural signals to combat the limited telemetry bandwidth when wireless communication with the external world is sought. The system has a multimodal processing capability to support a wide range of scenarios in real experimental conditions. A transmission link with rate-dependent compression and spike sorting strategy is shown to preserve information fidelity. At 32 channels sampled at 25 kHz, the power consumption of the system is 5.19 mW and has been implemented on a 5mm×5mm nano-FPGA, bringing its performance within the implantable power-size constraints for clinical applications.