Complementary Graphene-Ferroelectric Transistors (C-GFTs) as Synapses with Modulatable Plasticity for Supervised Learning

Novel complementary graphene-ferroelectric transistors (C-GFTs) based synapses are proposed and experimentally demonstrated for the first time. By exploiting the unique zero-bandgap property of graphene, GFT based synapses can be dynamically reconfigured between potentiative and depressive (PD) modes corresponding to hole-and electron dominated transport in graphene channels, respectively. Both modes demonstrate excellent linearity, small (2%) cycle-to-cycle variation and > 32 levels when used as synapses. By configurating the PD modes into a pair of C-GFTs, the hardware architecture of spiking neural networks (SNNs) can be substantially innovated, where the complicated circuitry previously required for supervised learning is now completely removed. With C-GFTs, a synapse footprint of 100 μm2 and a power consumption of 8 pJ/per operation are demonstrated in the MNIST learning task.