High-Performance Ferroelectric Field-Effect Transistors Based on Ultrathin Indium Oxide for Neuromorphic Computing

The emergence of artificial intelligence has revealed the limitations of traditional von Neumann computing systems in fulfilling the current computational requirements. In-memory computing (IMC) has been generally considered as a promising architecture to break the von Neumann bottleneck, where the FeFET is a strong candidate for developing IMC hardware, but remains challenging. In this work, we demonstrate a complementary metal oxide semiconductor-compatible In2O3 FeFET array for neuromorphic computing. The FeFETs exhibit excellent performance, including an ultrahigh on-off ratio (107), large memory window (>6 V), high endurance (107 cycles), long retention time (>10 years), low cycle-to-cycle variation (1.1%), high uniformity, and highly linear and symmetrical long-term potentiation (LTP)/long-term depression (LTD). Finally, we evaluate the performance of fabricated In2O3 FeFETs for image classification, and a high overall accuracy of 92.5% is achieved. These results suggest the great potential of In2O3 FeFET for constructing IMC hardware for neuromorphic computing.