150 nm × 200 nm Cross‐Point Hexagonal Boron Nitride‐Based Memristors

Abstract

systems, [2,3] due to their high performance, easy fabrication, and high-density integration. A memristor is a two-terminal metal/insulator/metal (MIM) cell that can change its conductivity depending on the electrical stresses previously applied between the two electrodes. [4] The memristors that exhibit the highest performances operate based on the formation and disruption of a conductive nanofilament (CNF) across the dielectric. [4] Depending on the metallic and insulating materials used, memristors with different performances can be fabricated. For example, Ti/HfO 2 /TiN memristors have shown low operation voltages (<0.3 V), [5] TiN/Ti/ HfO 2 /TiN memristors have shown fast switching speed (<1 ns), [6] TiN/Hf/HfO X / TiN memristors have shown low energy consumption (<0.1 pJ per transition), [7] Pt/Ta 2 O 5-X /TaO 2-X /Pt memristors have shown outstanding endurances (>10 12 cycles), [8] and Pt/TaOx/Pt memristors have shown long retention times (>10 years at 85 °C). [9] However, so far there is no materials combination able to meet all these performances simultaneously. [10] Furthermore, other desired properties of memristors like analog linear and symmetric switching or stable multilevel switching have never been reliably fabricated-some claims exist in the literature, but they do not prove the behaviors statistically providing device-to-device variability information. For these reasons, fabricating memristors using new materials that can meet the performances required is necessary. The introduction of 2D materials in memristive technologies has gained a substantial amount of interest because they can provide additional capabilities to the devices (e.g., flexibility and transparency). [11] For example, graphene has been used as interfacial layer to avoid the interactions between metallic electrodes and the resistive switching (RS) medium (e.g., oxygen vacancies migration from the insulator to electrodes, and metal ion penetration into the insulator). [12] Other 2D materials like graphene oxide (GO), [13-17] black phosphorous (BP), [18] molybdenum disulfide (MoS 2), [19,20] and hexagonal boron nitride (h-BN) [21] have been used as RS active medium, and the switching mechanisms have been attributed to the migration of intrinsic atomic species in the 2D material (e.g., O in GO, S in MoS 2 , and B in h-BN), and in some cases it is combined with the penetration of metal ions. [22] Unfortunately, 2D materials based memristors are still not being taken seriously because: i) most of the works just show proof-of-concept RS observations (e.g., no variability The introduction of 2D materials in the structure of memristors has been shown to provide the devices with enhanced flexibility and transparency. However, their use is still not well justified, as the electrical performance of 2D materials-based memristors is still behind that of transition metal oxide (TMO)-based memristors. This work presents the fabrication of metal/h-BN/ metal memristors with ultra-low power consumption that beat the previous record set by Au/HfO...

Wafer‐Scale 2D Hafnium Diselenide Based Memristor Crossbar Array for Energy‐Efficient Neural Network Hardware

“…Furthermore, a statistical analysis of V set , reset voltage ( V reset ), conductance of HRS and LRS are shown in Figure 3d,e, showing a comparable device variation to other 2D material‐based memristors. [ 52 ] In Figure 3d, an average V set of ≈0.7 V is measured, leading to a low power consumption. Figure 3e shows the cumulative probability distribution of HRS and LRS, and a RS ratio of ≈50 times is retained despite experiencing a spatial variation.…”

“…[62] Moreover, the endurance test at 85 °C is also shown in Figure S18, Supporting Information, indicating the stability under harsh operating conditions. [52,63] The sensitivity of offline classification to RS ratio is also investigated. Figure 4e shows an apparent improvement in recognition accuracy with increasing RS ratio as it could compensate for the non-zero HRS leakage current.…”

Variability and Yield in h‐BN‐Based Memristive Circuits: The Role of Each Type of Defect

“…The correct morphology of the h-BN has been confirmed via cross-sectional transmission electron microscopy (TEM) in some of our previous publication. [21,22] The samples made of monolayer h-BN exhibit the formation of cracks, and the cracked area ranges from ≈5.40% to ≈11.92% at different regions of the sample (see also Figure S1, Supporting Information); on the contrary, the samples made of multilayer h-BN show correct morphology free of cracks in large areas >1 mm 2 (Figure S2, Supporting Information). As an example, Figure S3 (Supporting Information) shows the optical microscopic image of one portion of a crossbar array of 100 × 100 devices.…”

Dielectric Breakdown in Single-Crystal Hexagonal Boron Nitride

“…Clearly there is an overlap of the breakdown regions once the separation reduces to ∼100 nm or less. For practical applications in 2D nanoelectronics, the spatial extent of the individual breakdown spot provides an estimate on the ultimate density of h-BN-based 2D transistors or memristors that could be reliability integrated, e.g., ∼10 9 transistors/cm 2 for devices with dimensions of 300 nm × 300 nm.…”