A Brain-Inspired In-Memory Computing System for Neuronal Communication via Memristive Circuits

Ji, Xiaofan; Dong, Zhekang; Lai, Chun Sing; Qi, Donglian

doi:10.1109/mcom.001.21664

Cited by 39 publications

(25 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Deep neural network layers make the extracted features more task-specific dependent, while shallow layers learn general features [ 16 , 17 ]. To fully extract feature information, the marginal feature extraction module and the conditional feature extraction module were designed.…”

Section: Methodsmentioning

confidence: 99%

Deep Transfer Network with Multi-Space Dynamic Distribution Adaptation for Bearing Fault Diagnosis

Zheng

Liu

et al. 2022

Entropy

View full text Add to dashboard Cite

Domain adaptation-based bearing fault diagnosis methods have recently received high attention. However, the extracted features in these methods fail to adequately represent fault information due to the versatility of the work scenario. Moreover, most existing adaptive methods attempt to align the feature space of domains by calculating the sum of marginal distribution distance and conditional distribution distance, without considering variable cross-domain diagnostic scenarios that provide significant cues for fault diagnosis. To address the above problems, we propose a deep convolutional multi-space dynamic distribution adaptation (DCMSDA) model, which consists of two core components: two feature extraction modules and a dynamic distribution adaptation module. Technically, a multi-space structure is proposed in the feature extraction module to fully extract fault features of the marginal distribution and conditional distribution. In addition, the dynamic distribution adaptation module utilizes different metrics to capture distribution discrepancies, as well as an adaptive coefficient to dynamically measure the alignment proportion in complex cross-domain scenarios. This study compares our method with other advanced methods, in detail. The experimental results show that the proposed method has excellent diagnosis performance and generalization performance. Furthermore, the results further demonstrate the effectiveness of each transfer module proposed in our model.

show abstract

Section: Methodsmentioning

confidence: 99%

Deep Transfer Network with Multi-Space Dynamic Distribution Adaptation for Bearing Fault Diagnosis

Zheng

Liu

et al. 2022

Entropy

View full text Add to dashboard Cite

show abstract

“…Meanwhile, the proposed circuit has advantages in terms of area overhead (achieving minimum value: 74.83μm 2 ) and power consumption (achieving minimum value: 100.99μW), which indicates that the proposed system is cost saving and energy-efficient. Furthermore, different with [4][5][6][7][8][9], this work provides a flexible scheme with fully hardware-implemented training strategy that can realize a specialized ANN and its variants, indicating a more universal application.…”

Section: Gradient Calculationmentioning

confidence: 99%

“…[3]. Memristors are two-terminal electronic devices that exhibit non-volatility, high density, long retention, and long endurance, which are potential candidates for neuromorphic computing [4]. Recently, memristor-based neuromorphic computing systems with different learning algorithms have been developed to realize specialized neural networks, such as Long Short-term Memory (LSTM) networks, Spiking Neural Networks (SNNs), Convolutional Neural Networks (CNNs), which can be extended to realize a variety of applications e.g., affective communication.…”

Section: Introductionmentioning

confidence: 99%

Design and Implementation of a Flexible Neuromorphic Computing System for Affective Communication via Memristive Circuits

Dong

Lai

et al. 2023

IEEE Commun. Mag.

View full text Add to dashboard Cite

Neuromorphic computing is expected to realize fast and energy-efficient artificial neural networks and address the inherent limitations of von Neumann architectures in dedicated communication applications. To realize this vision, we identify the existing challenges in neuromorphic computing and provide a specific solution from the perspectives of device, circuit, and system. At the device level, we fabricate a metal oxide-based memristor with high stability, low power, and good scalability, serving as the fundamental component of neuromorphic computing system. At the circuit level, the basic circuit units and necessary peripheral circuits are designed to realize efficient vector-matrix multiplication and different functions, including nonlinear activation operation, subtraction operation, added operation, etc. At the system level, a flexible neuromorphic computing system with a hardware-friendly training approach is proposed, it can perform affective communication with good trade-off between accuracy and time consumption. This study is expected to achieve the deep integration of nanotechnology, energy-efficient integrated circuits, and neuromorphic computing systems into communication applications.

show abstract

“…However, the long-time standing makes it unable to be used for vehicle real-time SOC estimation, which is often used for offline detection (Dong et al, 2021). In addition, the OCV-SOC curve of the lithium-ion battery has a flat area, and a slight change in voltage in this area will cause a large SOC error, which has high requirements for the accuracy of the voltage sampling circuit (Ji et al, 2022).…”

Section: Literature Reviewmentioning

confidence: 99%

State of Charge Estimation for Lithium-Ion Batteries Using Simple Recurrent Units and Unscented Kalman Filter

Zhang

Zhu

et al. 2022

Front. Energy Res.

View full text Add to dashboard Cite

Accurate estimation of the state of charge plays a very important role in ensuring the safe and effective operation of battery lithium-ion batteries and is one of the most important state parameters. However, the estimation method of state of charge has various limitations, so it is of great significance to improve the accuracy and calculation speed of the method. In this article, we propose an improved recurrent neural network model to estimate lithium-ion battery state of charge. Simple recurrent units are used to replace the traditional recurrent neural network basic unit or long short-term memory unit, and the computation speed is improved by implementing parallel processing. Finally, the prediction results of the model are fed into an unscented Kalman filter module to remove the interference of noise on the prediction. This article studies the prediction accuracy and speed of Samsung INR 18650-20R and INR 18650-25R under various ambient temperatures, initial state of charge values, and electric vehicle drive cycles. The results show that the proposed method can obtain accurate state of charge estimation results in the INR 18650-20R data set. For different temperatures and initial SOC, the root mean square error is less than 0.015 and 0.016, and the prediction speed is about 30% higher than that of long short-term memory. In the INR 18650-25R data set, for three different driving cycles, the root mean square error is less than 0.034, and the average test speed is about 2.7s, which proves the effectiveness of this method in estimating accuracy and speed.

show abstract

A Brain-Inspired In-Memory Computing System for Neuronal Communication via Memristive Circuits

Cited by 39 publications

References 12 publications

Deep Transfer Network with Multi-Space Dynamic Distribution Adaptation for Bearing Fault Diagnosis

Deep Transfer Network with Multi-Space Dynamic Distribution Adaptation for Bearing Fault Diagnosis

Design and Implementation of a Flexible Neuromorphic Computing System for Affective Communication via Memristive Circuits

State of Charge Estimation for Lithium-Ion Batteries Using Simple Recurrent Units and Unscented Kalman Filter

Contact Info

Product

Resources

About