Modeling spatio-temporal read voltages with complex distortions arising from the write and read mechanisms in flash memory devices is essential for the design of signal processing and coding algorithms. In this work, we propose Flash-Gen, a data-driven approach to generating flash memory read voltages in both space and time using conditional generative networks. This generative modeling method reconstructs read voltages from an individual memory cell based on the program levels of the cell and its surrounding cells, as well as the time stamp, in a time-efficient, resource-saving, and function-comprehensive manner. We evaluate the model over a range of time stamps using the read voltage distributions, the cell level error rates, and the relative frequency of errors for patterns most susceptible to inter-cell interference (ICI) effects. We propose a flash system optimization procedure, referred to as the Flash-Gen coding workflow, that leverages reconstructed read voltages for the development of error correction codes (ECCs) and constrained codes. Experimental results demonstrate that the model accurately captures the complex spatial and temporal features of the flash memory channel. Flash-Gen coding workflow can effectively address a range of important tasks, including threshold determination, coding performance estimation, and pattern characterization.

The goal of functional error correction is to preserve neural network performance when stored network weights are corrupted by noise. To achieve this goal, a selective protection (SP) scheme was proposed to optimally protect the functionally important bits in binary weight representations in a layer-dependent manner. Although it showed its effectiveness in image classification tasks on some relatively simple networks such as ResNet-18 and VGG-16, it becomes inadequate for emerging complex machine learning tasks generated from natural language processing and vision-language association domains. To solve this problem, we extend the SP scheme in three directions: task complexity, model complexity, and storage complexity. Extensions to complex natural language and vision-language tasks include text categorization and “zero-shot” textual classification of images. Extensions to more complex models with deeper block structures and attention mechanisms consist of Very Deep Convolutional Neural Network (VDCNN) and Contrastive Language-Image Pre-Training (CLIP) networks. Extensions to more complex storage configurations focus on distributed storage architectures to support model parallelism. Experimental results show that the optimized SP scheme preserves network performance in all of these settings. The results also provide insights into redundancy-performance tradeoffs, generalizability of SP across datasets and tasks, and robustness of partitioned network architectures.

Modeling spatio-temporal read voltages with complex distortions arising from the write and read mechanisms in flash memory devices is essential for the design of signal processing and coding algorithms. In this work, we propose a data-driven approach to modeling NAND flash memory read voltages in both space and time using conditional generative networks. This generative flash modeling (GFM) method reconstructs read voltages from an individual memory cell based on the program levels of the cell and its surrounding cells, as well as the time stamp. We evaluate the model over a range of time stamps using the cell read voltage distributions, the cell level error rates, and the relative frequency of errors for patterns most susceptible to inter-cell interference (ICI) effects. Experimental results demonstrate that the model accurately captures the complex spatial and temporal features of the flash memory channel.

The pivotal storage density win achieved by solid-state devices over magnetic devices in 2015 is a result of multiple innovations in physics, architecture, and signal processing. One of the most important innovations in that regard is enabling the storage of more than one bit per cell in the Flash device, i.e., having more than two charge levels per cell. Constrained coding is used in Flash devices to increase reliability via mitigating inter-cell interference that stems from charge propagation among cells. Recently, capacity-achieving constrained codes were introduced to serve that purpose in modern Flash devices, which have more than two levels per cell. While these codes result in minimal redundancy via exploiting the underlying physics, they result in non-negligible complexity increase and access speed limitation since pages cannot be read separately. In this paper, we suggest new constrained coding schemes that have low-complexity and preserve the desirable high access speed in modern Flash devices. The idea is to eliminate error-prone patterns by coding data either only on the left-most page (binary coding) or only on the two left-most pages (4-ary coding) while leaving data on all the remaining pages uncoded. Our coding schemes work for any number of levels$q \geq 4$per cell, offer systematic encoding and decoding, and are capacity-approaching. Since the proposed schemes enable the separation of pages, except the two left-most pages in the case of 4-ary coding, we refer to them as read-and-run (RR) constrained coding schemes as opposed to schemes adopting read-and-wait for other pages. The 4-ary RR coding scheme is introduced in order to limit the rate loss incurred by the binary RR coding schemes, and we show that our 4-ary RR coding scheme is also competitive when it comes to complexity and error propagation. We analyze the new RR coding schemes and discuss their impact on the probability of occurrence of different charge levels. We also demonstrate the performance improvement achieved via RR coding on a practical triple-level cell Flash device.

The pivotal storage density win achieved by solid-state devices over magnetic devices in 2015 is a result of multiple innovations in physics, architecture, and signal processing. One of the most important innovations in that regard is enabling the storage of more than one bit per cell in the Flash device, i.e., having more than two charge levels per cell. Constrained coding is used in Flash devices to increase reliability via mitigating inter-cell interference that stems from charge propagation among cells. Recently, capacity-achieving constrained codes were introduced to serve that purpose in modern Flash devices, which have more than two levels per cell. While these codes result in minimal redundancy via exploiting the underlying physics, they result in non-negligible complexity increase and access speed limitation since pages cannot be read separately. In this paper, we suggest new constrained coding schemes that have low-complexity and preserve the desirable high access speed in modern Flash devices. The idea is to eliminate error-prone patterns by coding data only on the left-most page while leaving data on all the remaining pages uncoded. Our coding schemes work for any number of levels per cell, offer systematic encoding and decoding, and are capacity-approaching. Since the proposed schemes enable the separation of pages, we refer to them as read-and-run (RR) constrained coding schemes as opposed to schemes adopting read-and-wait for other pages. We analyze the new RR coding schemes and discuss their impact on the probability of occurrence of different charge levels. We also demonstrate the performance improvement achieved via RR coding on a practical triple-level cell Flash device.

With the reduction in device size and the increase in cell bit-density, NAND flash memory suffers from larger inter-cell interference (ICI) and disturbance effects. Constrained coding can mitigate the ICI effects by avoiding problematic error-prone patterns, but designing powerful constrained codes requires a comprehensive understanding of the flash memory channel. Recently, we proposed a modeling approach using conditional generative networks to accurately capture the spatio-temporal characteristics of the read signals produced by arrays of flash memory cells under program/erase (P/E) cycling. In this paper, we introduce a novel machine learning framework for extending the generative modeling approach to the coded storage channel. To reduce the experimental overhead associated with collecting extensive measurements from constrained program/read data, we train the generative models via transferring knowledge from models pre-trained with pseudo-random data. This technique can accelerate the training process and improve model accuracy in reconstructing the read voltages induced by constrained input data throughout the flash memory lifetime. We analyze the quality of the model by comparing flash page bit error rates (BERs) derived from the generated and measured read voltage distributions. We envision that this machine learning framework will serve as a valuable tool in flash memory channel modeling to aid the design of stronger and more efficient coding schemes.

In this paper, we investigate the use of recurrent neural network (RNN)-based detection of magnetic recording channels with inter-symbol interference (ISI). We refer to the proposed detection method, which is intended for recording channels with partial-response equalization, as Partial-Response Neural Network (PR-NN). We train bi-directional gated recurrent units (bi-GRUs) to recover the ISI channel inputs from noisy channel output sequences and evaluate the network performance when applied to continuous, streaming data. The computational complexity of PR-NN during the evaluation process is comparable to that of a Viterbi detector. The recording system on which the experiments were conducted uses a rate-2/3, (1,7) runlength-limited (RLL) code with an E2PR4 partial-response channel target. Experimental results with ideal PR signals show that the performance of PR-NN detection approaches that of Viterbi detection in additive white gaussian noise (AWGN). Moreover, the PR-NN detector outperforms Viterbi detection and achieves the performance of Noise-Predictive Maximum Likelihood (NPML) detection in additive colored noise (ACN) at different channel densities. A PR-NN detector trained with both AWGN and ACN maintains the performance observed under separate training. Similarly, when trained with ACN corresponding to two different channel densities, PR-NN maintains its performance at both densities. Experiments confirm that this robustness is consistent over a wide range of signal-to-noise ratios (SNRs). Finally, PR-NN displays robust performance when applied to a more realistic magnetic recording channel with MMSE-equalized Lorentzian signals.

In this paper, we investigate the use of recurrent neural network (RNN)-based detection of magnetic recording channels with inter-symbol interference (ISI). We refer to the proposed detection method, which is intended for recording channels with partial-response equalization, as Partial-Response Neural Network (PR-NN). We train bi-directional gated recurrent units (bi-GRUs) to recover the ISI channel inputs from noisy channel output sequences and evaluate the network performance when applied to continuous, streaming data. The recording system on which the experiments were conducted uses a rate-2/3, (1,7) runlength-limited (RLL) code with an E2PR4 partial-response channel target. Experimental results with ideal PR signals show that the performance of PR-NN detection approaches that of Viterbi detection in additive white gaussian noise (AWGN). Moreover, the PR-NN detector outperforms Viterbi detection and achieves the performance of Noise-Predictive Maximum Likelihood (NPML) detection in additive colored noise (ACN).

This paper shows that the error probability of a querying policy for noisy 20 questions is upper bounded by the minimum Hamming distance of its querying matrix. Following this distance principle, sparse querying matrices with the row-column constraint are constructed for the scenarios, where only limited areas can be detected at each querying round. It demonstrates that the row-column constraint promises a large minimum distance to ensure low error probability of queries. Moreover, the proposed sparse matrices with random block coding provide unequal error-protection capability to further improve the querying accuracy under the scenarios of detecting half of the areas. Simulation results verify that the quantized mean squared errors of our proposed policies outperform those of the existing policies under the above scenarios.