Abstract
Forming gas annealing (FGA) is applied to HfO(x) ferroelectric tunnel junction (FTJ) synaptic devices to passivate defects and reduce trap-assisted-tunneling (TAT). Without FGA, TAT caused by defects in metal-ferroelectric-insulator-semiconductor (MFIS) FTJ stack dominates the conduction mechanism in FTJs and results in no memory window (MW). The reduction of defects or TAT after FGA reveals the effect of polarization switching on the FTJ performance. Consequently, linear/symmetric potentiation and depression (P/D) characteristics of FTJ after FGA with stable repeatability are obtained. Owing to the FGA-induced linearity and symmetricity of P/D, a learning accuracy of approximately 90% is achieved via pattern recognition simulations utilizing HfO(x) FTJ crossbar.