Article
Chemistry, Multidisciplinary
Anna N. Matsukatova, Aleksandr I. Iliasov, Kristina E. Nikiruy, Elena Kukueva, Aleksandr L. Vasiliev, Boris Goncharov, Aleksandr Sitnikov, Maxim L. Zanaveskin, Aleksandr S. Bugaev, Vyacheslav A. Demin, Vladimir V. Rylkov, Andrey Emelyanov
Summary: The paper proposes a hybrid CNN, utilizing a hardware fixed pre-trained feature extractor and a trainable software classifier. The hardware part is implemented using passive crossbar arrays of memristors. Experimental results show that the performance of the hybrid CNN is comparable to other memristor-based systems, while requiring significantly fewer trainable parameters. It also exhibits robustness to variations in memristor characteristics.
Article
Chemistry, Analytical
Jiyong An, Seokjin Oh, Tien Van Nguyen, Kyeong-Sik Min
Summary: This paper proposes a new crossbar training scheme that optimizes the defect map size and neural network performance. By dividing the columns of the memristor crossbar into different groups and combining synapse-aware and neuron-aware training methods, the proposed scheme can improve network performance while minimizing hardware burden.
Article
Engineering, Electrical & Electronic
Lixing Huang, Hongqi Yu, Changlin Chen, Jie Peng, Jietao Diao, Hongshan Nie, Zhiwei Li, Haijun Liu
Summary: Memristor-based convolutional neural networks (CNNs) have been extensively studied in the field of edge computing due to their high integration density and powerful processing capability. However, the low yield of memristor array and the variation in memristance pose limitations for the widespread application of memristor-based CNNs. To address this issue, a training strategy is proposed to enhance the robustness of memristor-based binarized neural networks for embedded applications.
SEMICONDUCTOR SCIENCE AND TECHNOLOGY
(2022)
Article
Materials Science, Multidisciplinary
T. Guo, K. Pan, B. Sun, L. Wei, Y. Yan, Y. N. Zhou, Y. A. Wu
Summary: The development of the first adjustable LIF neuron using novel memristor-coupled capacitors was achieved, along with the use of genetic algorithm to detach the entanglement of capacitive and memristive effects. This method can be generalized to other entangled physical behaviors, facilitating the development of novel circuits.
MATERIALS TODAY ADVANCES
(2021)
Article
Chemistry, Multidisciplinary
Xinyi Li, Yanan Zhong, Hang Chen, Jianshi Tang, Xiaojian Zheng, Wen Sun, Yang Li, Dong Wu, Bin Gao, Xiaolin Hu, He Qian, Huaqiang Wu
Summary: This study utilizes transition metal oxide-based memristors as artificial dendrites and spike-firing soma to construct dendritic neuron units, achieving high-efficiency spatial-temporal information processing. A hardware-implemented dendritic neural network improves accuracy for human motion recognition and exhibits a 1000x advantage in power efficiency compared to a graphics processing unit.
ADVANCED MATERIALS
(2023)
Article
Engineering, Electrical & Electronic
Sachin Maheshwari, Alexander Serb, Christos Papavassiliou, Themistoklis Prodromakis
Summary: In the pursuit of low power, bio-inspired computation, both memristive and memcapacitive-based Artificial Neural Networks (ANN) have gained increasing attention for hardware implementation. Taking a step further, regenerative capacitive neural networks, with the use of adiabatic computing and 'memimpedace' elements, offer a promising path for even lower energy consumption. This research proposes an artificial neuron with adiabatic synapse capacitors and demonstrates significant energy savings.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS I-REGULAR PAPERS
(2022)
Article
Mathematics
Tal Rozen, Moshe Kimhi, Brian Chmiel, Avi Mendelson, Chaim Baskin
Summary: This paper proposes a new binarization method that achieves a bimodal distribution of network weights through kurtosis regularization. Preserving the weight distribution during binarization training significantly reduces the generalization error of the BNN. Experimental results demonstrate that this new method outperforms current state-of-the-art schemes on CIFAR-10 and ImageNet.
Review
Materials Science, Multidisciplinary
Hui Chen, Huilin Li, Ting Ma, Shuangshuang Han, Qiuping Zhao
Summary: As the boom of data storage and processing, brain-inspired computing provides an effective approach to solve the current problem. Various emerging materials and devices have been reported to promote the development of neuromorphic computing. Herein, we mainly review the progress of these brain functions mimicked by neuromorphic devices, concentrating on synapse, neurons, and intelligent behaviors, and present some challenges and prospects related to neuromorphic devices.
SCIENCE AND TECHNOLOGY OF ADVANCED MATERIALS
(2023)
Article
Engineering, Electrical & Electronic
Quang-Manh Duong, Quang-Kien Trinh, Van-Tinh Nguyen, Dinh-Ha Dao, Duy-Manh Luong, Van-Phuc Hoang, Longyang Lin, John Deepu
Summary: This paper proposes a charge-based integrate-and-fire (IF) circuit for in-memory binary spiking neural networks (BSNNs). The circuit can perform addition and subtraction operations, which makes it compatible with in-memory XNOR-based synapses for implementing the BSNN processing core. A framework is developed to evaluate the circuit design by considering its imperfections effects in system-level simulation. The simulation results show that the proposed design achieves a performance of 5.10 fJ/synapse and an 82.01% classification accuracy for CIFAR-10 under process variation.
INTERNATIONAL JOURNAL OF CIRCUIT THEORY AND APPLICATIONS
(2023)
Article
Chemistry, Multidisciplinary
Jianmin Zeng, Xinhui Chen, Shuzhi Liu, Qilai Chen, Gang Liu
Summary: In this study, an artificial synapse based on an organic memristor with adjustable conductance states was developed. The memristor exhibited long-term synaptic plasticity and was utilized in a neural network for handwritten digit recognition with high accuracy.
Review
Neurosciences
Dmitry Ivanov, Aleksandr Chezhegov, Mikhail Kiselev, Andrey Grunin, Denis Larionov
Summary: This article discusses the limitations of modern artificial intelligence systems based on von Neumann architecture and classical neural networks compared to the mammalian brain, and ways to overcome them through neuromorphic AI projects. It also presents the principle of classifying neuromorphic AI systems based on the brain features they use, and discusses the prospects of using new memristor element base in neuromorphic applications.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Chemistry, Multidisciplinary
Tao Guo, Baizhou Zhang, Xiyang Wang, Yi Xiao, Bai Sun, Y. Norman Zhou, Yimin A. Wu
Summary: Traditional machine vision is faced with the challenges of redundant sensing data, bulky structures, and high energy consumption. However, biological-inspired neuromorphic systems offer a promising solution for compact and energy-efficient machine vision. Multifunctional optoelectronics that enable multispectrum sensitivity and neuromorphic computing play a vital role in achieving this goal.
ADVANCED FUNCTIONAL MATERIALS
(2023)
Article
Engineering, Electrical & Electronic
Huiyuan Liu, Xiaojian Zhu, Zhecheng Guo, Ri He, Xinze Li, Qihao Sun, Xiaoyu Ye, Cui Sun, Yu Tian, Run-Wei Li
Summary: This study presents a memristor that can mimic the burst-firing features of biological neurons by generating periodic voltage oscillation groups. The burst frequency can be adjusted and used for burst frequency coding. The artificial neural system based on bursting neurons achieves high recognition accuracy for Fashion-MNIST tasks.
ACS APPLIED ELECTRONIC MATERIALS
(2023)
Article
Materials Science, Multidisciplinary
Qi Liu, Song Gao, Yang Li, Wenjing Yue, Chunwei Zhang, Hao Kan, Guozhen Shen
Summary: This article introduces a WO3/HfO2 heterojunction-based memristor with extraordinary resistive switching behaviors and neuromorphic characteristics. The mechanism behind the electrical performances of this device is studied, and a multilayer perceptron neural network constructed based on the memristor model is explored to enhance recognition accuracy. The proposed memristor contributes to promoting the development of high-density storage and neuromorphic computing technology.
ADVANCED MATERIALS TECHNOLOGIES
(2023)
Article
Chemistry, Physical
Li Zhang, Zhenhua Tang, Junlin Fang, Xiujuan Jiang, Yan-Ping Jiang, Qi-Jun Sun, Jing-Min Fan, Xin-Gui Tang, Gaokuo Zhong
Summary: Artificial neural network-based computing has the potential to overcome the limitations of conventional computers and has a wide range of applications. By using NiO/Cu2O memristors to emulate biological synapses, the recognition accuracy of an artificial neural network based on synaptic weight modulation reached up to 96.84% on average, demonstrating the potential of artificial synapses in artificial intelligence systems.
APPLIED SURFACE SCIENCE
(2022)
Article
Engineering, Electrical & Electronic
Haijun Liu, Shengyang Sun, Jianjun Liu, Qingjiang Li, Jietao Diao, Zhiwei Li
Summary: The BMVTBP architecture combines the advantages of low-precision memristive devices and VNN to achieve MAC operations on interval data with good identification performance.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2021)
Article
Neurosciences
Shiqing Zhang, Hui Xu, Zhiwei Li, Sen Liu, Bing Song, Qingjiang Li
Summary: Ovonic threshold switch (OTS) has been widely studied in neuromorphic computing due to its high-density synapse array support, but a simple and complete model for device simulation and integrated circuit design has been lacking. In this study, a compact physical model of OTS based on the Poole-Frenkel effect and thermal dissipation effect was developed for the first time, showing good agreement with experimental results and offering insights into device performance optimization.
FRONTIERS IN NEUROSCIENCE
(2021)
Article
Engineering, Electrical & Electronic
Yufei Zhang, Hui Xu, Zhiwei Li, Yi Sun, Hongqi Yu, Changlin Chen
Summary: In the design of memristive chips, memristor models play an important role. However, existing models are either too complex to achieve satisfactory simulation speed or too simplified to accurately reflect device characteristics. This article proposes a PieceWise linear (PWL) memristor model with well-balanced simulation speed and accuracy, suitable for large-scale arrays.
IEEE TRANSACTIONS ON ELECTRON DEVICES
(2022)
Article
Engineering, Electrical & Electronic
Lixing Huang, Hongqi Yu, Changlin Chen, Jie Peng, Jietao Diao, Hongshan Nie, Zhiwei Li, Haijun Liu
Summary: Memristor-based convolutional neural networks (CNNs) have been extensively studied in the field of edge computing due to their high integration density and powerful processing capability. However, the low yield of memristor array and the variation in memristance pose limitations for the widespread application of memristor-based CNNs. To address this issue, a training strategy is proposed to enhance the robustness of memristor-based binarized neural networks for embedded applications.
SEMICONDUCTOR SCIENCE AND TECHNOLOGY
(2022)
Article
Computer Science, Hardware & Architecture
Junwei Zeng, Nuo Xu, Yabo Chen, Chenglong Huang, Zhiwei Li, Liang Fang
Summary: Traditional CMOS-based von-Neumann computer architecture faces the issue of memory wall, which limits data processing efficiency and consumes energy due to the speed mismatch between processor and memory. Recently, novel in-memory computing paradigms using non-volatile memories have emerged as promising solutions. In this study, we propose a new in-memory computing unit based on a memory array with magnetoelectric spin-orbit logic (MESO) device, which allows for several logic operations within the memory array. The proposed design shows significant improvement in storage delay and logic delay compared to other spintronics-based devices.
ACM TRANSACTIONS ON DESIGN AUTOMATION OF ELECTRONIC SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Jiwei Li, Hui Xu, Sheng-Yang Sun, Nan Li, Qingjiang Li, Zhiwei Li, Haijun Liu
Summary: This article introduces an improved training algorithm for multilayer memristive spiking neural networks (MSNN) that supports in situ learning in hardware through three methods spontaneously regulating hidden neurons and updating weights in situ. Experimental results demonstrate that the proposed MSNN achieves high recognition accuracy and robustness, performing well against nonideal factors.
IEEE TRANSACTIONS ON COGNITIVE AND DEVELOPMENTAL SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Xingzhi Fu, Qingjiang Li, Weihe Wang, Hui Xu, Yinan Wang, Wei Wang, Hongqi Yu, Zhiwei Li
Summary: This brief introduces a high-speed memristor-based N-bit ripple carry adder (RCA) with a special design to facilitate implementation in 1T1R arrays. The proposed RCA uses Memristor-Aided LoGIC (MAGIC) to calculate and store outputs in parallel, resulting in a significant reduction in delay compared to existing memristive RCAs.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS
(2022)
Article
Nanoscience & Nanotechnology
Weihe Wang, Yinan Wang, Zhiwei Li, Xingzhi Fu, Mingxin Deng, Xiaojuan Liu, Qingjiang Li, Hui Xu
Summary: This paper proposes a high-reliable 2T2R synaptic structure for constructing a 4-bit MN-ADC with high-speed and accurate conversion capability, which is perfectly compatible with 1T1R crossbar arrays.
Article
Engineering, Electrical & Electronic
Weihe Wang, Zhiwei Li, Xingzhi Fu, Yinan Wang, Qingjiang Li
Summary: This paper proposes a pipelined Hopfield neural network architecture for memristor-based analog-to-digital converter (ADC), which allows self-adaptive weight tuning. The training algorithm is modified to reduce complexity and make it hardware-friendly. The synapse matrix can adapt to the crossbar array. The proposed architecture achieves good performance in simulation but is limited by the comparator in experimental performance.
IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Yinan Wang, Hakan Johansson, Mingxin Deng, Zhiwei Li
Summary: This paper comprehensively analyzes and compares two different timing-mismatch compensation strategies for two-channel time-interleaved analog-to-digital converters. The paper introduces a novel compensation structure that achieves a remarkably greater spurious-free dynamic range (SFDR) compared to existing structures. Theoretical derivations and simulations validate the proposed structures.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2022)
Article
Automation & Control Systems
Zhiwei Li, Hongchang Long, Xi Zhu, Yinan Wang, Haijun Liu, Qingjiang Li, Nuo Xu, Hui Xu
Summary: This paper proposes a method of error detection and correction (EDC) based on memristor-based stateful logic operation, which improves accuracy by leveraging the concept of redundancy, and demonstrates the feasibility of the proposed method in practical devices.
ADVANCED INTELLIGENT SYSTEMS
(2022)
Proceedings Paper
Engineering, Electrical & Electronic
Hongchang Long, Jietao Diao, Xi Zhu, Zhiwei Li, Haijun Liu
Summary: In-memory computing based on memristor technology has attracted significant interest as a potential solution to break the data-transfer bottleneck in von Neumann computer architecture. Using memristor-based logic gates, such as material implication (IMP), offers a way to achieve various Boolean functions. By deducing formulas and designing voltage application schemes, different Boolean functions can be realized effectively with the help of these logic gates' structures.
2021 IEEE INTERNATIONAL SYMPOSIUM ON CIRCUITS AND SYSTEMS (ISCAS)
(2021)
Proceedings Paper
Engineering, Electrical & Electronic
Xi Zhu, Hui Xu, Hongchang Long, Qingjiang Li, Zhiwei Li, Haijun Liu, Yinan Wang
Summary: Although memristive stateful logic is a promising candidate for energy-efficient edge computing, reliability remains a challenge. This study analyzes the reliability of memristive stateful logic and proposes an n-modular redundancy method for error correction, which efficiently increases successful operation rate.
2021 5TH IEEE ELECTRON DEVICES TECHNOLOGY & MANUFACTURING CONFERENCE (EDTM)
(2021)