Resource Optimized Processor for Real-Time Neural Activity Monitoring

Closing the loop between living tissues and electronics is at the basis of the treatment of numerous pathologies or disabilities with prosthetic devices: artificial pancreas, neuroprostheses, BCI, etc. Biosignals are acquired and processed to detect a signature that finally controls actuators (insulin delivery pump, electrical stimulator, etc.). The in vivo application of such a paradigm implies severe constraints on speed and consumption at all loop processing stages. In the research described in this paper, we optimized computation in the first stage of an acquisition system dedicated to closed-loop living/artificial experiments. The described functions are wavelet-based spike detection and slow signal filtering. We focus here on the optimization of these algorithms to reduce hardware resources while keeping a strong constraint on the computation time, and easy scalability for massive multichannel recordings. The architecture is implemented on FPGA for prototyping and evaluation with living cells recorded on a 60 multi-electrode array. We describe the minimal requirements of the algorithms, present the computation architecture, and the required hardware resources, as well as the evolution of these resources depending on the number of recording channels.