Coexistence of volatile and non-volatile characteristics in SiO2/CoOx memristor for in-materia reservoir computing

In the era of big data, energy-efficient reservoir computing systems (RCSs) demonstrate potential to drive advancements in neuromorphic computing. While previous research has predominantly focused on synaptic memristor devices for the physical reservoirs of RCSs, the development of fully hardware-based RCSs remains crucial for achieving simplified processing and reduced power consumption. This study developed a fully hardware-based RCS using Au/SiO2/CoOx/ITO memristor devices, exhibiting both short- and long-term memory characteristics. These memristor devices achieved a high rectification ratio (>3 x 10(4)) and excellent endurance (similar to 12,000 cycles), enabling scalable crossbar array implementation. We also demonstrated synaptic functions, including paired pulse facilitation and spike-rate-dependent plasticity. After confirming the suitability of the memristor devices for the hardware readout layers of RCSs, we combined 16 distinct reservoir states in the reservoir layer with linear long-term potentiation and long-term depression in the readout layer and developed a fully hardware-based RCS. This RCS equipped with Au/SiO2/CoOx/ITO devices, which can be fabricated cost-effectively using the solgel method, achieved 93.39 % accuracy in MNIST image classification, demonstrating the potential of Au/SiO2/CoOx/ITO memristors for applications in fully hardware-based RCSs.