Coexistence of threshold and memory switching characteristics of ALD HfO₂ memristor synaptic arrays for energy-efficient neuromorphic computing.

We demonstrate inherent biorealistic synaptic plasticity functions in the Pt/n-ZnO/SiO 2-x /Pt heterostructures, where the n-ZnO semiconductor is geometrically cone-shaped in the size of a few nanometers. The synaptic functions were achieved within a two-terminal, electroforming-free, and low-power rectifying diode-like resistive switching device. The important rate-dependent synaptic functions, such as the nonlinear transient conduction behavior, short-and long-term plasticity, paired-pulse facilitation, spike-rate-dependent plasticity and sliding threshold effect, were investigated in a single device. These characteristics closely mimic the memory and learning functions of those in biosynapses, where frequency-dependent identical spiking operations are mostly taking place, and we emulate these characteristics in the "Learning-Forgetting-Relearning" synaptic behavior. The switching dynamics in the cone-shaped n-ZnO semiconductor are correlated with the transport mechanism along the grain boundaries of the charged ion species, namely, oxygen vacancies and charged oxygen. The diffusion and generation/recombination of these defects have specific time scales of self-decay by virtue of the asymmetric profile of the n-ZnO cone defects. Finally, the essential biorealistic synaptic plasticity functions were discovered for the perspectives of dynamic/adaptive electronic synapse implementations in hardware-based neuromorphic computing.

We report the dependence of the thickness of amorphous boron nitride (a-BN) on the characteristics of conductive bridge random access memory (CBRAM) structured with the Ag/a-BN/Pt stacking sequence. The a-BN thin film layers of three different thicknesses of 5.5, 11, and 21.5 nm were prepared by the sputtering deposition. Depending on the thickness of the a-BN layer, the devices are found to be in either low-resistance state (LRS) or high-resistance state (HRS) prior to any consecutive switching cycle. All devices with 5.5 nm thick a-BN switching layer are in LRS as the pristine state, while devices with 21.5 nm thick a-BN layer are found to be in HRS as the pristine state. To attain reliable switching cycles, initial RESET and electroforming process are necessarily required for the devices with 5.5 and 21.5 nm thick a-BN layer, respectively. However, the devices with the a-BN layer of thickness between 5.5 and 21.5 nm in pristine states are in either HRS or LRS. This dependence of the a-BN thickness on different resistance states in the pristine state can be explained by in situ Ag diffusion during its sputter deposition to form a top electrode on the a-BN layer. Our finding shows a detailed investigation and a deep understanding of the switching mechanism of Ag/a-BN/Pt CBRAM devices with respect to different a-BN thicknesses for the future computing system.

No abstract

No abstract