Source-linked AI summary
Multibit memory operation of metal-oxide bi-layer memristors
Spyros Stathopoulos, Ali Khiat, Maria Trapatseli, Simone Cortese, Alexandrou Serb, Ilia Valov, Themis Prodromakis
TL;DR
TiO2-only devices showed at most 10 observed resistive states, while the study evaluates metal-oxide bilayers for multibit operation. Using a current-compliance-free pulsing protocol, it reports 47 stable states, corresponding to 5.5 bits per cell.
Problem
The multistate capability of TiO2-only devices was limited to a maximum of 10 observed states, motivating evaluation of barrier-layer bilayers.
Source: p. 2Method
The study evaluates metal-oxide bilayers and uses a current-compliance-free pulsing protocol to program stable multilevel ReRAM states.
Source: p. 3Results
47 stable resistive states yielded 5.5 bits per cell, with selected states showing excellent retention over 8 hours.
Source: p. 2Takeaways & Limitations
Bilayer-based memristors may provide a viable technology path for next-generation nonvolatile memory and neuromorphic applications.
Source: p. 3
Abstract
from arXiv · showhide
In this work, we evaluate a multitude of metal-oxide bi-layers and demonstrate the benefits from increased memory stability via multibit memory operation. We introduce a programming methodology that allows for operating metal-oxide memristive devices as multibit memory elements with highly packed yet clearly discernible memory states. We finally demonstrate a 5.5-bit memory cell (47 resistive states) with excellent retention and power consumption performance. This paves the way for neuromorphic and non-volatile memory applications.
Figure captions
The captions describe multibit operation of AlxOy/TiO2 devices, including closely packed, individually discernible resistive states, retention measurements, and the energy required to reach higher states. They also outline comparisons across barrier-layer combinations and between TiO2-only and bilayer devices.
- TiO2-only and AlxOy/TiO2 bilayer devices are shown with typical bipolar switching under alternating-polarity 100 ns voltage ramps from 1 to 2 V in 200 mV steps.The figure also includes a device micrograph and schematics of the single-layer and bilayer stacks, both with platinum top and bottom electrodes.Source: Figure 1
- The established resistive states are closely packed and individually discernible, with select states measured for 8-hour retention and higher states requiring exponentially increasing switching energy.The caption also describes cumulative probability distributions for the states and energy measurements for the states shown in that distribution.Source: Figure captions · p. 7, Figure 3
- Starting from ∼21.5 kΩ, sequential 100 ns SET pulses at 2 V program specific states; a train of 100×100 ns RESET pulses at −2 V returns the device to baseline.The caption describes flushing the device to baseline before switching into a different state.Source: Figure 3
- The multibit evaluation compares barrier-layer combinations by attainable resistive-state count and final-state-to-baseline resistance ratio.State assessment uses a 2σ confidence interval; individual assessed states appear in supplementary Figs. S6 and S7.Source: Figure 2 · p. 8
Supplementary Information
Supplementary results describe the multistate evaluation procedure and show that bilayer stacks generally improve TiO2-only devices. AlxOy/TiO2 provides 46 states with improved dynamic range and linearity, while attainable-state counts depend on the confidence bound used.
Optimised pulsing protocol for a TiO2-only device
- Alternating-polarity 1–3 V, 1 µs pulses with 100 mV steps improve TiO2-only device stability until worst-case low- and high-resistance switching windows no longer overlap.The improved stability requires more energy because the pulses are longer and higher-voltage.Source: Figure 2
State evaluation algorithm
- The evaluation algorithm identifies switching polarity, drives the device to a stable resistance, then tests new states against a 2σ separation criterion until voltage or monotonicity limits are reached.It uses at least 50 points for stability evaluation and terminates at maximum voltage or when the state sequence becomes non-monotonic.Source: Figure 4
Effect of the confidence bounds
- Using 3σ confidence bounds instead of 2σ roughly halves registered AlxOy/TiO2 states, from 47 to 23; the passage describes 95% confidence as sufficient for practical adjacent-state discrimination.The figure identifies 3σ as 99.7% and 2σ as 95%.Source: Figure 5
Short-term retention for all bilayer combinations
- Except for SiO2, bilayer combinations improve both attainable-state count and established-state stability over TiO2-only devices in short-term retention measurements.Final-state measurements used 50 × 100 ns pulses with 20 ms intervals.Source: Figure 6
Multistate evaluation for different bi-layer combinations
- 46 resistive states, overall linearity, and increased dynamic range make AlxOy/TiO2 the most promising tested stack for granular, predictable multibit storage.All studied bilayer combinations increase the number of states over TiO2, but only AlxOy/TiO2, WO3/TiO2, and HfO2/TiO2 also increase dynamic range.Source: Supplementary Information · p. 10, Figure 7