Energy consumption of SOMN-based physical reservoir computing

Determine the energy consumption required to execute real-world tasks using Self-Organising Memristive Networks (SOMNs) as physical reservoirs, in order to assess their practical viability for edge and embedded machine intelligence applications.

Background

The paper surveys physical reservoir computing using Self-Organising Memristive Networks (SOMNs), highlighting their nonlinearity, fading memory, and multi-terminal readout capabilities. These properties have enabled a range of tasks, including speech and audio recognition, time-series prediction, and pattern recognition, using both nanowire and nanoparticle networks.

Despite these advances, the authors explicitly note that key practical aspects remain unresolved, particularly energy usage for real-world tasks. Since SOMNs are proposed as energy-efficient substrates for edge intelligence, quantifying energy consumption is essential to benchmark them against alternative neuromorphic and conventional approaches.