Ratio-based multi-level resistive memory cells

Miguel Ángel Lastras-Montaño, Osvaldo Del Pozo-Zamudio, Lev Glebsky, Meiran Zhao, Huaqiang Wu, Kwang‐Ting ChengScientific 2021 · DOI

TL;DR

This paper introduces a novel "ratio-based encoding" (RatioBE) for multi-level resistive memory cells (MLCs), which significantly improves data reliability and/or reduces programming energy compared to traditional resistance-based encoding (ReBE). By using the ratio of resistances from two devices instead of a single device's resistance, RatioBE dramatically lowers the bit error probability (BEP), enabling more robust and efficient memory systems.

What Problem Does This Paper Address?

Resistive Random-Access Memory (ReRAM) is a promising non-volatile memory technology, but its widespread adoption is hindered by significant device-to-device and cycle-to-cycle variations. These variations lead to large distributions in the memory states' resistances, which is particularly problematic for Multi-Level Cells (MLCs) that aim to store multiple bits per cell by relying on precise resistance levels. Traditional resistance-based encoding (ReBE) struggles to overcome these variations, leading to high bit error probabilities (BEPs). While various methods exist to mitigate these issues, they often involve complex device or circuit improvements, or iterative programming schemes that increase programming time and energy. This paper addresses the fundamental challenge of encoding information reliably in the presence of device variability, proposing a more efficient and robust encoding scheme.

Methods

The core innovation is the generalization of ratio-based encoding (RatioBE) from single-level to multi-level memory cells. Instead of using the absolute resistance of a single device to represent a state (ReBE), RatioBE employs two serially connected resistive-switching devices configured as a voltage divider. The memory state is then determined by the normalized output voltage of this divider. The authors derive a closed-form expression for the BEP for both RatioBE and ReBE, allowing for a theoretical comparison. They also experimentally validate their findings using various ReRAM devices, comparing the BEP and programming effort (PE) of both encoding schemes. The PE is related to the number of programming iterations required to achieve a certain level of device variability.

Datasets

The paper does not explicitly mention or introduce new benchmark datasets. Instead, it relies on experimental data obtained from fabricated ReRAM devices and commercially available devices. These include:

  • A 1 Kb 1T1R hafnium-oxide ReRAM array.
  • Commercially available tungsten-based devices.
  • Multi-level titanium-based devices.

The experimental results are used to validate the theoretical models and demonstrate the practical advantages of RatioBE.

Key Insights and Ideas

  • RatioBE offers a fundamental advantage in combating variability: By using a ratio of resistances, the encoding becomes inherently more robust to absolute resistance variations. An error only occurs if the ratio of resistances crosses a threshold, which is a less probable event than a single resistance value crossing a threshold.
  • Theoretical BEP reduction is significant: The derived closed-form expression shows that for a given ReBE system with BEP $x$, RatioBE can achieve a BEP of $x^2$ in the best case and $x\sqrt{2}$ in the worst case. This implies a substantial improvement, especially for lower BEPs.
  • Trade-offs between BEP, capacity, and PE are manageable: RatioBE allows for flexibility. Researchers can either achieve significantly lower BEPs, increase memory capacity (more bits per cell), or reduce programming time/energy while maintaining a comparable BEP.
  • RatioBE is complementary to existing improvements: The proposed encoding scheme is orthogonal to device and circuit-level improvements, meaning it can benefit from and be combined with other efforts to enhance ReRAM performance.
  • Experimental validation confirms theoretical predictions: The authors demonstrate substantial improvements in BEP and/or programming efficiency across different device types, aligning with their theoretical models.

Key Results

  • BEP Reduction: RatioBE can achieve up to 3 orders of magnitude lower BEP compared to ReBE on the same devices.
  • Programming Efficiency: RatioBE can reduce programming time and energy by 5-10x while achieving the same BEP as ReBE.
  • Memory Capacity: For a given BEP and PE, RatioBE can offer 20-40% higher memory capacity.
  • Theoretical BEP Relationship: For a ReBE system with BEP $10^{-x}$, RatioBE can achieve a BEP of $10^{-2x}$ to $10^{-\sqrt{2}x}$.
  • Experimental BEP Improvement: In one experiment, RatioBE achieved a 1000x reduction in BEP compared to ReBE.
  • Programming Effort Reduction: Experiments showed that RatioBE could achieve similar reliability with only 20% of the programming time/energy compared to ReBE.

Limitations and Future Work

  • Device Variability Modeling: While the paper models resistance states as log-normally distributed, experimental data sometimes uses normal distributions for current-based states. The authors note that the estimation of BEP assuming normally distributed states was within ±25% of the log-normal estimation, and truncation effects were minimal (<7%).
  • Experimental Complexity: Implementing RatioBE requires two devices per cell, which might increase the cell's physical footprint compared to a single-device ReBE cell, although the authors suggest that in a 1T1R architecture, the transistor often dominates the cell size.
  • Future Work: The authors believe RatioBE is essential for tackling the intrinsic variability of ReRAM and other resistance-switching devices, paving the way for next-generation high-performance and high-capacity non-volatile memories. Further exploration into optimizing RatioBE for specific applications and device technologies is implied.

Summary updated

Abstract

Ratio-based encoding has recently been proposed for single-level resistive memory cells, in which the resistance ratio of a pair of resistance-switching devices, rather than the resistance of a single device (i.e. resistance-based encoding), is used for encoding single-bit information, which significantly reduces the bit error probability. Generalizing this concept for multi-level cells, we propose a ratio-based information encoding mechanism and demonstrate its advantages over the resistance-based encoding for designing multi-level memory systems. We derive a closed-form expression for the bit error probability of ratio-based and resistance-based encodings as a function of the number of levels of the memory cell, the variance of the distribution of the resistive states, and the ON/OFF ratio of the resistive device, from which we prove that for a multi-level memory system using resistance-based encoding with bit error probability x, its corresponding bit error probability using ratio-based encoding will be reduced to [Formula: see text] at the best case and [Formula: see text] at the worst case. We experimentally validated these findings on multiple resistance-switching devices and show that, compared to the resistance-based encoding on the same resistive devices, our approach achieves up to 3 orders of magnitude lower bit error probability, or alternatively it could reduce the cell's programming time and programming energy by up 5-10[Formula: see text], while achieving the same bit error probability.
BibTeX
@inproceedings{lastrasmontano2021ratio,
  author = {Miguel Ángel Lastras-Montaño and Osvaldo Del Pozo-Zamudio and Lev Glebsky and Meiran Zhao and Huaqiang Wu and Kwang‐Ting Cheng},
  title = {{Ratio-based multi-level resistive memory cells}},
  booktitle = {Scientific},
  year = {2021},
  doi = {10.1038/s41598-020-80121-7},
}
Figure