Hybrid digital-neuromorphic architecture integration for low-power applications

Innovative computing paradigms based on machine learning are revolutionizing the world of IoT, where devices that were once seen as end-nodes connected to the internet to acquire and transmit data now are required to also perform computation. This poses new challenges for developers, since edge devices need to satisfy strict requirements in terms of energy efficiency and limited computational resources. Spiking Neural Networks (SNN), thanks to their capability of emulating the data processing of biological neurons, are promising candidates for low-power edge applications. In this work-in-progress paper we would like to explore the co-existence of a traditional processor with a SNN accelerator through the SPI protocol used for configuring the SNN. Respectively PULPissimo, a RISC-V-based microcontroller, and ReckOn, an open source neuromorphic processor. Later on, we present our next steps towards integrating ReckOn as a co-processor inside PULPissimo.