Impact of area ratio in oxide-semiconductor FeFET synapses with a metal-ferroelectric-metal-insulator-semiconductor structure for neuromorphic application

Abstract Neuromorphic computing that enables parallel computation and adaptive learning process has the potential to overcome the traditional von Neumann bottleneck. To build a neuromorphic system, synapse is deemed as one of the most essential elements. In this study, adopting a metal-ferroelectric-metal-insulate-semiconductor (MFMIS) structure, we demonstrated a hafnia-based oxide-semiconductor-channel ferroelectric field-effect transistor (OS-FeFET) as a synapse with multilevel characteristics, and investigated the impact of area ratio ( AR ) between the MOS capacitor ( A MOS ) and the ferroelectric capacitor ( A FE ) on the memory and synaptic performances through the simulation. The present results show that not only the memory window but also the linearity of synaptic weight update and dynamic range could be significantly improved as the AR increases. Furthermore, an artificial neural network (ANN) based on these OS-FeFET synapses was constructed, which could achieve 85.3% accuracy in recognizing the handwritten digits. This study may provide insights for future OS-FeFET synapse design for neuromorphic application.