Energy Efficient Hybrid Reservoir Computing Using Hfn.5Zrn.5O2 Ferroelectric Thin-Film Transistors with an Integrated Optically and Electrically Synaptic Functions

Abstract

This study introduces an ultralow power hybrid reservoir computing (HRC) system employing an indium gallium zinc oxide (IGZO)/Hf0.5Zr0.5O2 (HZO)-based ferroelectric thin-film transistor (FeTFT) for neuromorphic applications. The proposed FeTFT system integrates volatile and nonvolatile functionalities, respectively driven by optical and electrical stimuli, to emulate short-term and long-term synaptic behaviors. Leveraging persistent photoconductivity in the IGZO channel under optical excitation, the FeTFT exhibits dynamic reservoir characteristics, while HZO-induced ferroelectric polarization enables robust long-term memory for the readout layer. Experimental results demonstrate enhanced energy efficiency with a power consumption of approximate to 22 pW per device and distinct separation of 4- and 5-bit reservoir states. This system achieves competitive accuracies of 90.48% and 88.23% for Modified National Institute of Standards and Technology (MNIST) and fashion MNIST datasets, respectively, surpassing state-of-the-art hardware-based implementations. By consolidating reservoir and readout layers within a single device, this study advances the scalability and feasibility of next-generation neuromorphic computing systems. Furthermore, the implementation of HRC leveraging optical and electrical pulses presents promising prospects for applications involving visual neuron functionalities.