MPSoCs for real-time neural signal decoding: A low-power ASIP-based implementation

Abstract

In this paper we target the design of a dedicated low-power computing platform for neuroprosthetic applications. The system must be capable of decoding the information encoded in neural signals, to extract the patients’ motion intention. To this aim, a highly-portable and reliable integrated processing device is required. However, a commonly acknowledged design methodology, to be used in such kind of design cases, is still not available in literature. In this work, we propose and assess the adoption of the MPSoC paradigm as a prospective solution. We present a design-case of a custom MPSoC integrated solution, implementing an on-line neural signal decoding algorithm. The proposed system executes parallel software tasks onto customized ASIP processing cores. Experimental results, obtained by placement- and activity-aware power evaluations carried out using an industrial 40 nm technology node as a reference, assess that the performance and power-related features of the designed architecture are compliant with the implantability constraints and with the battery lifetime required for real-life use. Moreover, besides the effectiveness of the proposed solution, this paper demonstrates also that custom heterogeneous MPSoCs can successfully challenge ultra-low power bio-medical signal processing problem.