Homo-layer hafnia-based memristor with large analog switching window

Abstract

The fast development of high-accuracy neuromorphic computing requires stable analog memristors. While filamentary memory switching is very common in binary oxides, their resistive switching usually involves abrupt changes due to the rupture or reformation of metallic filaments. In this work, we designed a memristor consisting of dual-layer HfOy/HfOx, with different concentrations of oxygen vacancies (y > x). During the electroforming process, both the migration of existing oxygen vacancies in HfOx and the generation of new oxygen vacancies in HfOy occur simultaneously, leaving a semiconducting part close to the HfOy/HfOx interface. The resulting filament is not metallic as a whole, as revealed by first principles calculations. Such a device demonstrates excellent switching uniformity as well as highly gradual resistance change, ideal for neuromorphic computing. Through fine tuning of the filament structure, the device achieves low variation, high speed, gradual SET and RESET processes, and hundreds of stable multi-level state behaviors. The long-term synaptic plasticity was further achieved, showing good linearity and large analog switching window (ΔG as high as 487.5 μS). This works affords a route toward a gradual resistance change in oxide-based memristors.

Unveiling the Potential of HfO2/WS2 Bilayer Films: Robust Analog Switching and Synaptic Emulation for Advanced Memory and Neuromorphic Computing

“…The conductance increases and decreased intensity patterns are shown in Figure d-f throughout 300 synaptic pulses, thus explaining the dependability of weight regulation. The findings show that during cycle repetition, memristors with similar pulse amplitudes perform effectively in managing conductance and display a much wider dynamic range of conductance than those with nonidentical pulse amplitudes Figure f displays how the conductance changes of LTP and LTD are affected by the number of normalized pulses, as demonstrated by the mathematical model expressed by eqs and G = ( false( G L R S ∝ − G H R S ∝ false) × w + G H R S ∝ ) 1 / ∝ ⁣ i f ∝ ≠ 0 G = G H R S × ( G L R S / G H R S ) w ⁣ i f ∝ = 0 where G HRS and G HRS are conductance values in the LRS and HRS, respectively, α is a fitting parameter indicating the linearity of the weight updating in the potentiation (α p ) or depression (α d ) process, and w is an internal variable (from 0 to 1) that increases/decreases during the potentiation/depression process.…”

“…The findings show that during cycle repetition, memristors with similar pulse amplitudes perform effectively in managing conductance and display a much wider dynamic range of conductance than those with nonidentical pulse amplitudes. 13 where G HRS and G HRS are conductance values in the LRS and HRS, respectively, α is a fitting parameter indicating the linearity of the weight updating in the potentiation (α p ) or depression (α d ) process, and w is an internal variable (from 0 to 1) that increases/decreases during the potentiation/ depression process. When the device undergoes linear conductance modulation (ideal case), α = 1, as shown by the black lines in Figure 4g-i.…”

“…Multilevel switching in filament-based RRAMs enables the storage of data on numerous layers. This is achieved by modifying the reset–stop voltage or managing current compliance during the set operation . The interface growth/shrinkage and electron trapping/detrapping at the Schottky barrier have significant impacts on multilevel switching .…”

“…This is achieved by modifying the reset− stop voltage or managing current compliance during the set operation. 13 The interface growth/shrinkage and electron trapping/detrapping at the Schottky barrier have significant impacts on multilevel switching. 14 In neuromorphic computing, memristors have been extensively researched as a way to emulate synaptic processes, such as long-term potentiation (LTP), long-term depression (LTD), paired-pulse facilitation (PPF), spike rate-dependent plasticity (SRDP), and spiketiming-dependent plasticity (STDP).…”

“…The findings show that during cycle repetition, memristors with similar pulse amplitudes perform effectively in managing conductance and display a much wider dynamic range of conductance than those with nonidentical pulse amplitudes. 13…”