Multilayer Reservoir Computing Based on Ferroelectric α-In2Se3 for Hierarchical Information Processing

Dynamic physical systems such as reservoir computing (RC) architectures show a great prospect in temporal information processing, whereas hierarchical information processing capability is still lacking due to the absence of advanced multilayer reservoir elements. Here, a stackable reservoir system is constructed based on ferroelectric alpha-In2Se3 devices with voltage input and output, which is realized by dynamic voltage division between a ferroelec-tric field-effect transistor and a planar device and therefore allows the reservoirs to cascade, enabling multilayer RC. Fast Fourier transformation analysis shows high-harmonic generation in the first layer as a result of inherent non-linearity of the reservoir, and progressive low-pass filtering effect is realized where higher-frequency components are progressively filtered in deeper-layer RCs. Time-series prediction and waveform classification tasks are also demonstrated, serving as evidence for the memory capacity and computing capability of the deep reservoir architecture. This work can provide a promising pathway in exploiting emerging 2D materials and dynamics for advanced neuromorphic computing architectures.

Automated summary

A stackable reservoir system based on ferroelectric alpha-In2Se3 devices was constructed, enabling multilayer RC with high-low pass filtering effects. The study also demonstrated the potential of time-series prediction and waveform classification tasks, serving as evidence for the memory capacity and computing capability of the deep reservoir architecture.