Exploring the Impact of Random Telegraph Noise-Induced Accuracy Loss on Resistive RAM-Based Deep Neural Network

Du, Yide; Jing, Linglin; Fang, Hui; Chen, Hai-Bao; Cai, Yimao; Wang, Runsheng; Zhang, Chenglong; Ji, Zhigang

doi:10.1109/ted.2020.3002736

Cited by 18 publications

(17 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Read noise, often also called RTN, refers to random jumps of the read current in a frequency range between 10 0 and 10 4 Hz. It has been attributed to either the charging/activation and discharging/deactivation of traps [21,41], or the random redistribution of oxygen vacancies in the oxide layer [33]. The amount of read noise was found to be stronger for higher resistance states [28,33,41], as well as for increased temperatures and read voltages [19].…”

Section: Read Instabilitymentioning

confidence: 99%

“…Read disturb describes the fact that over multiple readouts of a VCM device its resistance might drift from its initial value or even completely switch to the opposite state [16][17][18]. Read noise describes the fact that when reading out a VCM device the read-out current is not constant but will show random jumps whose height depends on the resistance value [19][20][21][22]. This noise is often referred to as random telegraph noise (RTN) [23][24][25].…”

Section: Introductionmentioning

confidence: 99%

“…Most analyses so far have focused on the impact of programming variability, read disturb and read noise on the performance of neural networks [21,28]. In [29], the impact of different non-idealities such as line parasitics, read disturb, and read noise on the IMPLY logic was studied.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Reliability aspects of binary vector-matrix-multiplications using ReRAM devices

Bengel

Mohr

Wiefels

et al. 2022

Neuromorph. Comput. Eng.

View full text Add to dashboard Cite

Computation-In-Memory (CIM) using memristive devices is a promising approach to overcome the performance limitations of conventional computing architectures introduced by the von Neumann bottleneck which are also known as memory wall and power wall. It has been shown that accelerators based on memristive devices can deliver higher energy efficiencies and data throughputs when compared with conventional architectures. In the vast multitude of memristive devices, bipolar resistive switches (BRS) based on the valence change mechanism (VCM) are particularly interesting due to their low power operation, non-volatility, high integration density and their CMOS compatibility. While a wide range of possible applications is considered, many of them such as artificial neural networks heavily rely on Vector-Matrix-Multiplications (VMMs) as a mathematical operation. These VMMs are made up of large numbers of Multiplication and Accumulation (MAC) operations. The MAC operation can be realised using memristive devices in an analog fashion using Ohm’s law and Kirchhoff’s law. However, VCM devices exhibit a range of non-idealities, affecting the VMM performance, which in turn impacts the overall accuracy of the application. Those non-idealities can be classified into time-independent (programming variability) and timedependent (read disturb and read noise). Additionally, peripheral circuits such as Analog to Digital Converters (ADCs) can introduce errors during the digitalization. In this work, we experimentally and theoretically investigate the impact of deviceand circuit-level effects on the VMM in a VCM crossbars. Our analysis shows that the variability of the Low Resistive State (LRS) plays a key role and that reading in the RESET direction should be favored to reading in the SET direction.

show abstract

Section: Read Instabilitymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Reliability aspects of binary vector-matrix-multiplications using ReRAM devices

Bengel

Mohr

Wiefels

et al. 2022

Neuromorph. Comput. Eng.

View full text Add to dashboard Cite

show abstract

“…When we read it from the EMT cell, the output may become w + ∆w instead of w. Here ∆w is a fluctuating amplitude of that memory cell. Because of this instability, in-memory deep learning using EMT especially in an analog manner may make incorrect classification [10]. This can severely limit its application in the real world.…”

Section: Introductionmentioning

confidence: 99%

“…• The stochastic parameter s ijkl is a random coefficient indicating whether the memory cell storing weight w ik is at l-th state when it is sampled and multiplied with the input vector x j . We denote matrix [s ijkl ] as S ij , shown in Equation (10). S ij can be considered as a part of fluctuation data S.…”

mentioning

confidence: 99%

Optimizing for In-memory Deep Learning with Emerging Memory Technology

Wang¹,

Luo²,

Goh³

et al. 2021

Preprint

View full text Add to dashboard Cite

In-memory deep learning computes neural network models where they are stored, thus avoiding long distance communication between memory and computation units, resulting in considerable savings in energy and time. In-memory deep learning has already demonstrated orders of magnitude higher performance density and energy efficiency. The use of emerging memory technology promises to increase the gains in density, energy, and performance even further. However, emerging memory technology is intrinsically unstable, resulting in random fluctuations of data reads. This can translate to non-negligible accuracy loss, potentially nullifying the gains. In this paper, we propose three optimization techniques that can mathematically overcome the instability problem of emerging memory technology. They can improve the accuracy of the in-memory deep learning model while maximizing its energy efficiency. Experiments show that our solution can fully recover most models' state-of-the-art accuracy, and achieves at least an order of magnitude higher energy efficiency than the state-of-the-art.

show abstract

Optimization of Random Telegraph Noise Characteristics in Memristor for True Random Number Generator

Song

Kim

Hwang

et al. 2023

Advanced Intelligent Systems

View full text Add to dashboard Cite

Memristor devices can be utilized for various computing applications, and stochastic computing is one of them. The intrinsic stochastic characteristics of the memristor cause unpredictable current fluctuations by the capture and emission of electrons in a trap site. Herein, a true random number generator (TRNG) using the random telegraph noise (RTN) of the memristor as an entropy source is proposed. TiOx/Al2O3 memristors are fabricated, and the capture time probability of the RTN characteristics is modulated to 50% with varying read‐voltage and device conductance state. In addition, the TRNG operations with the RTN current signals as entropy sources are experimentally demonstrated with a customized breadboard, and the randomness is verified with the National Institute of Standards and Technology (NIST) tests.

show abstract

Exploring the Impact of Random Telegraph Noise-Induced Accuracy Loss on Resistive RAM-Based Deep Neural Network

Cited by 18 publications

References 26 publications

Reliability aspects of binary vector-matrix-multiplications using ReRAM devices

Reliability aspects of binary vector-matrix-multiplications using ReRAM devices

Optimizing for In-memory Deep Learning with Emerging Memory Technology

Optimization of Random Telegraph Noise Characteristics in Memristor for True Random Number Generator

Contact Info

Product

Resources

About