MgO/HZO Based Ferroelectric Tunnel Junctions for Neuromorphic Computing Applications

In this work, we fabricate a ferroelectric tunnel junction (FTJ) device with W/MgO/HZO/TiN structure. The effect of MgO thickness on the FTJ properties of HZO films is studied. Device with 0.5 nm thick MgO insulating layer and annealing at 600 °C for 20 s exhibits a large memory window of about 30 and stable endurance of at least one million cycles with multilevel states. For pulse measurements, this device shows excellent nonlinearities of 0.1 and 0.32 for potentiation(P) and depression(D), respectively. The conductance data of P and D are input into simulated Hopfield neural network model for training to learn
$10\times10$
-pixel size images. Such model recognizes the input images to reach an accuracy of 100 % in only 19 iterations. This result demonstrates that our FTJ device has high potential working as artificial synapse for the neuromorphic computing application in the future.