InGaZnO-based synaptic transistor with embedded ZnO charge-trapping layer for reservoir computing

In this study, we design an IGZO/ZnO/IGZO-based synaptic transistor to implement robust reservoir computing. Short-term memory characteristics are achieved using the charge trapping and detrapping effects of the ZnO layer. We verify excellent cell-to-cell and cycle-to-cycle uniformity regarding the memory characteristics of the device. Moreover, various synaptic behaviors, including short-term potentiation, short-term depression (STD), excitatory postsynaptic currents (EPSC), and paired-pulse facilitation (PPF) are emulated to check the suitability of neuromorphic properties. Finally, reservoir computing trained on the modified National Institute of Standards and Technology database dataset is presented for temporal data learning. As a physical reservoir, the device can achieve 16 different using 4 bits depending on the applied pulse stream. The results include a confusion matrix covering all recognition scenarios, with an average recognition accuracy of 93.87%, closely approaching the theoretical recognition accuracy of 94.1%. This study sheds light on a computational framework for physical reservoir computing by reducing the training cost.