[Paper Review] Increasing Flash Memory Lifetime by Dynamic Voltage Allocation for Constant Mutual Information

Tsung‐Yi Chen, Adam R. Williamson+1 more|arXiv (Cornell University)|Mar 18, 2014

Advanced Data Storage Technologies20 references3 citations

TL;DR

This paper proposes a dynamic voltage allocation scheme in flash memory that maintains constant mutual information by adjusting read voltages based on wear-out, extending 4-level MLC flash lifetime by nearly doubling it. The method uses soft information from multiple reads to estimate noise distributions and adapt voltage levels in real time.

ABSTRACT

The read channel in Flash memory systems degrades over time because the Fowler-Nordheim tunneling used to apply charge to the floating gate eventually compromises the integrity of the cell because of tunnel oxide degradation. While degradation is commonly measured in the number of program/erase cycles experienced by a cell, the degradation is proportional to the number of electrons forced into the floating gate and later released by the erasing process. By managing the amount of charge written to the floating gate to maintain a constant read-channel mutual information, Flash lifetime can be extended. This paper proposes an overall system approach based on information theory to extend the lifetime of a flash memory device. Using the instantaneous storage capacity of a noisy flash memory channel, our approach allocates the read voltage of flash cell dynamically as it wears out gradually over time. A practical estimation of the instantaneous capacity is also proposed based on soft information via multiple reads of the memory cells.

Motivation & Objective

• Address the degradation of flash memory due to Fowler-Nordheim tunneling and charge injection over program/erase cycles.
• Overcome the limitation of fixed voltage thresholds that lead to reduced reliability and shorter device lifetime as cells wear out.
• Maintain constant mutual information between input and output of the flash channel throughout the device’s operational life.
• Develop a practical, real-time method to estimate wear-out and retention noise using existing multi-read soft information from LDPC decoding.

Proposed method

• Model the flash memory channel as a time-varying noisy system with input-dependent Gaussian and Laplace noise components.
• Use the instantaneous channel capacity as a metric to determine optimal voltage levels that preserve constant mutual information.
• Dynamically adjust read voltage thresholds based on estimated noise distribution from histogram-based soft information obtained via multiple reads.
• Apply numerical optimization to minimize total accumulated charge (V_acc) while maintaining target capacity, thus extending lifetime.
• Leverage existing multi-read capability in modern flash systems to estimate threshold voltage distributions without additional hardware.
• Use quantized soft information from LDPC decoding as input to estimate mean and variance shifts due to wear and retention loss.

Experimental results

Research questions

• RQ1 Can dynamic adjustment of read voltage thresholds maintain constant mutual information in a flash memory channel as it degrades over time?
• RQ2 How can wear-out and retention noise be accurately estimated in real time using only existing multi-read soft information?
• RQ3 What is the maximum lifetime extension achievable through dynamic voltage allocation under realistic flash channel models?
• RQ4 How does the proposed method compare to static voltage threshold schemes in terms of charge accumulation and reliability?

Key findings

• The dynamic voltage allocation method extends the lifetime of a 4-level MLC flash memory by nearly doubling it under the chosen parameters.
• The approach maintains constant mutual information by adapting read voltage levels in response to estimated wear-out and retention noise.
• Histograms derived from multiple reads provide sufficient accuracy to estimate noise distribution and guide voltage level adjustments.
• The method is compatible with existing flash architectures, as it reuses soft information already generated for LDPC decoding.
• The technique effectively mitigates mean-shift effects from retention loss by identifying and correcting for bin shifts in threshold voltage distributions.
• The framework is generalizable to other flash memory channel models beyond the specific Gaussian and Laplace noise model used in the evaluation.