Enhancement of Multi-Timescale Reservoir Computing Based on Homogeneous Hardware Platform with Controllable Ionic Dynamics

As one kind of recurrent neural network, the reservoir computing (RC) can be well employed to process temporal information. Inspired by the emerging neuromorphic devices, we proposed a homogeneous hardware platform by adopting the solid electrochemical memtransistors. By tuning the base gate voltages, the ionic dynamics can be well controlled to meet the multi-timescale temporal information requirements. By testing the handwritten digit-recognition task, the accuracies can get improved from 77.744% to 82.17% when using the adjustable RC systems.