Evolving Memristive Reservoir

Memristive reservoirs, with their dynamic plasticity, nanosize, and energy efficiency, have gained significant attention in various research fields. However, their deterministic hardware implementation poses challenges for hardware reservoir adaptation. Traditional evolutionary algorithms for evolving reservoirs often overlook the circuit scalability and feasibility of memristive reservoirs. In this work, we introduce an evolvable memristive reservoir circuit based on reconfigurable memristive units (RMUs) that can adaptively evolve for different tasks, directly evolving the configuration signals of memristors to avoid device variance. We also propose a scalable algorithm for evolving this reconfigurable memristive reservoir circuit, ensuring circuit validity according to circuit laws and sparse topology for scalability and feasibility during evolution. We demonstrate the feasibility and superiority of our proposed evolvable memristive reservoir circuit through experiments on a wave generation task, six prediction tasks, and one classification task.