Nonvolatile Electrochemical Random‐Access Memory under Short Circuit

Abstract

gate memory, [6] phase-change memory (PCM), [7][8][9] ferroelectric transistors, [10,11] resistive redox memory (ReRAM), [12][13][14][15][16][17] and magnetic tunnel junction memory. [18] Unfortunately, none of these technologies have been able to meet all the demands of in-memory computing.Electrochemical random-access memory, or ECRAM, is a promising recently developed device for analog inmemory computing. ECRAM stores analog information states by electrochemically shuttling guest dopant ions like lithium ions, protons, or oxygen vacancies between two mixed-conducting host materials. Like ReRAM, ECRAM also stores and switches information through ion migration and the resulting change in valence and electronic conductivity; however, unlike ReRAM, ECRAM contains a solid electrolyte, enabling accurate, linear, and deterministic changes in the resistance state. Since research in ECRAM has dramatically increased from 2017, [21,22] it has met many of the requirements for analog in-memory computing, including linear and symmetric among hundreds of analog states; low read and write currents, voltages, and energies; parallel weight updates; [23,35] CMOS-compatible materials with back-end-of-the-line (BEOL) thermal budgets; [27,28,47] and scaled sub-micrometer sized devices. [24,28] Recently, submicrosecond switching times have been shown using protonsbased ECRAM into polymers, [23,48] MXenes, [34] and tungsten oxides. [28] These developments make ECRAM highly promising for on-line training, which aims to identify the correct analog resistance states (or weights) of a network using the training set of pre-classified data. An even larger application of in-memory computing is inference, which uses a pretrained network to classify new data. Inference is not only simpler to conduct in analog, but is more ubiquitous because all distributed devices, including ones with very limited energy budgets, would use inference to classify new data.An overarching challenge that has precluded ECRAM from inference accelerators is insufficient retention, generally only several hours at room temperature, [22,[25][26][27] and much lower than the ten years at 85 °C benchmark for ReRAM, PCM, and floating gate memory. Moreover, nearly all retention studies utilize an electronic switch that keeps the gate and channel electrodes under "open circuit," while this switch can provide several hours of retention in larger devices >10 µm, leakage currents in the switch will preclude retention times in scaled Electrochemical random-access memory (ECRAM) is a recently developed and highly promising analog resistive memory element for in-memory computing. One longstanding challenge of ECRAM is attaining retention time beyond a few hours. This short retention has precluded ECRAM from being considered for inference classification in deep neural networks, which is likely the largest opportunity for in-memory computing. In this work, an ECRAM cell with orders of magnitude longer retention than previously achieved is developed, and which is anticipated to excee...

Differentiating Oxygen Exchange Reaction Mechanisms across Phase Boundaries

“…Another potential area of impact might be the design of materials for emerging electrochemical ionic synaptic devices, , where the coupled ionic and electronic transfer (CIET) governs the operation speed. Inducing phase transition has been shown to be an effective method for improving the performance of the devices, which makes it essential to understand the link between phase transition and interfacial CIET. Therefore, we conclude that our findings pave the way toward a deeper understanding of phase-specific kinetics and reaction mechanisms for a variety of electrochemical interfaces.…”

Abisko: Deep codesign of an architecture for spiking neural networks using novel neuromorphic materials

“…Previously, we demonstrated a retention of ≈3 h for TiO x ECRAM that utilizes YSZ solid electrolyte (Li et al, 2020b). More recently, we demonstrated in collaboration with Y. Li’s group at the University of Michigan that a retention of ≈10 years is feasible for WO 3-x /YSZ ECRAM, as shown in Figure 20 (Kim et al, 2022).

Figure 20.ECRAM device.

…”

“…(d) Comparison of retention times of WO 3-x ECRAM with filament-based ReRAM and past TiO x -based ECRAM. (Kim et al, 2022). Copyright 2022, Wiley.…”

Electrochemical Ionic Synapses with Mg2+ as the Working Ion

“…This is attributed to slow oxidation of the channel by oxygen in air, even when HfO 2 passivation layers are used to protect the device from the atmosphere. [ 35 ] Furthermore, O‐EIS devices typically must be operated either using high voltages [ 26,28,31 ] or at elevated temperatures [ 29,35 ] to overcome the high activation energies for O 2− ion migration.…”