Voltage-operating reservoir computing utilizing ferroelectric-FET source follower configurations

Abstract We investigate FeFET reservoir computing (RC) utilizing a voltage readout scheme instead of the conventional current readout scheme to reduce the power consumption of an RC system and expand the range of applications. The voltage-output RC operates with time-series Vout data of a source follower of a FeFET (FeSF), to effectively utilize temporal memory as well as nonlinearity originating from ferroelectric polarization dynamics. We show that this RC system can solve several nonlinear tasks including time-series data prediction. Furthermore, to achieve higher computational performance in voltage-output FeSF RC, an FET is inserted between the source of the FeFET and the ground, which can operate as a current source to mitigate the slow pull-down operation of voltage-output FeSF RC. We also propose an RC system utilizing voltage responses from complementary working FeSFs with original and inverted gate input patterns to demonstrate high RC performance.