Morphological Reservoir Computing Hardware

Abstract

In the recent years, Reservoir Computing arises as an emerging machine-learning technique that is highly suitable for time-series processing. In this work, we propose the implementation of reservoir computing systems in hardware via morphological neurons which make use of tropical algebra concepts that allow us to reduce the area cost in the neural synapses. The main consequence of using tropical algebra is that synapses multipliers are substituted by adders, with lower hardware requirements. The proposed design is synthesized in a Field-Programmable Gate Array (FPGA) and benchmarked against a time-series prediction task. The current approach achieves significant savings in terms of power and hardware, as well as an appreciable higher precision if compared to classical reservoir systems.