检索结果-南通市图书馆

Co-optimization strategy between array operation and weight mapping for flash computing arrays to achieve high computing efficiency and accuracy

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2023年第2期66卷 289-290页

作者： Guihai YU Peng HUANG Runze HAN Lixia HAN Xiaoyan LIU Jinfeng KANG Institute of Microelectronics Peking University

Dear editor,Flash memory is a promising candidate to realize in-memory computing due to its mature fabrication technology [1]. Various efforts to implement neural networks using flash computing arrays have been reported [2, 3]. To realize higher computing efficiency, it is an effective method to activate more word lines(WL) at the same time for flash computing arrays [4]. When increasing the activated WL,

关键词： Mapping

Design space for stabilized negative capacitance in HfO2 ferroelectric-dielectric stacks based on phase field simulation

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2021年第2期64卷 192-201页

作者： Pengying CHANG Gang DU Xiaoyan LIU Institute of Microelectronics Peking University

Polarization evolution with space and time in HfObased metal-ferroelectric-insulator-metal(MFIM) structure is studied based on the phase field model by self-consistently solving the two-dimensional time-dependent Ginzburg-Landau and Poisson equations. Through examining the domain wall and electrostatic(depolarization) energies compared with the negative ferroelectric(FE) anisotropy energy, the correlation between the domain pattern(phase transition) and negative capacitance(NC) effect is revealed for different structure parameters and material properties, including FE thickness and gradient coefficient,dielectric permittivity and thickness, and operation frequency of applied voltage. The design space for stabilized NC with(near) hysteresis-free operation accompanied with voltage amplification is limited by phase transition, implying the potentials and limitations of FE HfOfor energy-efficient steep-slope devices.

关键词： ferroelectric polarization dynamics hysteresis negative capacitance depolarization field

Monocular 3D gaze estimation using feature discretization and attention mechanism

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Optoelectronics Letters 2023年第5期19卷 301-306页

作者： SHA Tong SUN Jinglin PUN Siohang LIU Yu School of microelectronics Tianjin UniversityTianjin 300072China Institute of Microelectronics University of MacaoMacao 999078China

Gaze estimation has become an important field of image and information processing.Estimating gaze from full-face images using convolutional neural network(CNN) has achieved fine accuracy.However,estimating gaze from eye images is very challenging due to the less information contained in eye images than in full-face images,and it’s still vital since eye-image-based methods have wider applications.In this paper,we propose the discretization-gaze network(DGaze-Net) to optimize monocular three-dimensional(3D) gaze estimation accuracy by feature discretization and attention mechanism.The gaze predictor of DGaze-Net is optimized based on feature discretization.By discretizing the gaze angle into K bins,a classification constraint is added to the gaze predictor.In the gaze predictor,the gaze angle is pre-applied with a binned classification before regressing with the real gaze angle to improve gaze estimation accuracy.In addition,the attention mechanism is applied to the backbone to enhance the ability to extract eye features related to gaze.The proposed method is validated on three gaze datasets and achieves encouraging gaze estimation accuracy.

关键词： estimation mechanism discretization

维普期刊数据库

The Principle and Progress of Dynamically Reconfigurable Computing Technologies

在线全文

学校读者我要写书评

暂无评论

Chinese Journal of Electronics 2020年第4期29卷 595-607页

作者： WEI Shaojun LU Yan’an Institute of Microelectronics Tsinghua University

With the emergence of new applications and the increasing cost of new semiconductor manufacturing technology, high energy-efficiency and flexibility are both critical for processors.Dynamically reconfigurable computing architecture, which has the both characteristics, is one of the most promising architectures for future processors.It has shown notable advantages in many important application fields, compared to conventional architectures.However, the fundamental reasons for the characteristics have never been deeply discussed in previous papers.This paper analyzes the reasons from the perspective of design methodology.The technologies of dynamically reconfigurable computing are continually evolving, and some cheering results of application have been achieved.This paper summarizes the latest progress in key technologies and provides an introduction of the application achievements.

关键词： Dynamically reconfigurable computing Principle Progress

Freely switching between ferroelectric and resistive switching in Hf0.5Zr0.5O2 films and its application on high accuracy on-chip deep neural networks

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2023年第2期66卷 267-276页

作者： Pengfei JIANG Kunran XU Jie YU Yannan XU Peng YUAN Yuan WANG Yuting CHEN Yaxin DING Shuxian LV Zhiwei DANG Tiancheng GONG Yang YANG Yan WANG Qing LUO Key Laboratory of microelectronics Devices and Integrated Technology Institute of MicroelectronicsChinese Academy of Sciences University of Chinese Academy of Sciences School of microelectronics Xidian University Peng Cheng Laboratory

The Hf0.5Zr0.5O2(HZO)-based ferroelectric field-effect transistor(Fe FET) synapse is a promising candidate for at-scale deep neural network(DNN) applications, because of its high symmetry, great accuracy and fast operation speed. However, the degradation of the remanent polarization(Pr) over time caused by the depolarization field has not been effectively resolved, greatly affecting the accuracy of the trained DNN. In this study, we demonstrate a ferroelectric(FE)-resistive switching(RS) switchable synapse using the FE mode for high-speed weight training and the RS mode for stable weight storage to overcome accuracy degradation. The FE-RS hybrid characteristic is accomplished by an HZO-based metal-ferroelectricmetal(MFM) capacitor with asymmetric electrodes, and the best FE endurance, as well as the most reliable RS behavior, is demonstrated by testing several electrodes materials. High memory windows are achieved in both FE and RS modes. Through this design, excellent accuracy is maintained over time, as verified by network simulation.

关键词： Hf 0.5 Zr 0.5 O 2 films ferroelectric resistive switching accuracy on-chip DNN

Body Bias Dependence of Bias Temperature Instability(BTI)in Bulk FinFET Technology

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Energy & Environmental Materials 2022年第4期5卷 1200-1203页

作者： Jiayang Zhang Zirui Wang Runsheng Wang Zixuan Sun Ru Huang Institute of Microelectronics Peking UniversityBeijing 100871China

In this article,the body bias dependence of the bias temperature instability(BTI)in bulk FinFETs is experimentally studied,under different test conditions for the first time.In contrast to the traditional understanding that changing body bias has little impact on BTI degradation in FinFETs due to its weak body effect,it is observed that it actually has non-negligible impacts.And a forward body bias(FBB)can reduce the BTI degradation in FinFETs,which is opposite with the trend in planar devices.The underlying physics is found due to the trade-off between two competing factors.The results are helpful for understanding and modeling reliability in FinFETs.

关键词： bias temperature instability(BTI) body effect FinFET reliability

维普期刊数据库

Circuit design of RRAM-based neuromorphic hardware systems for classification and modified Hebbian learning

在线全文

学校读者我要写书评

暂无评论

Science China(Information Sciences) 2019年第6期62卷 141-159页

作者： Yuning JIANG Peng HUANG Zheng ZHOU Jinfeng KANG Institute of Microelectronics Peking University

This paper proposes a solution to the learning of neuromorphic hardware systems based on metal-oxide resistive random access memory(RRAM) arrays, which are used as binary electronic synapses.A modified Hebbian learning method is developed to update the binary synaptic weights, and mixed-signal circuits are designed to implement the proposed learning method. The circuits are verified by SPICE, and systematic simulations are also conducted to verify the capability of the neuromorphic system to process relatively large databases. The results show that the system presents high processing speed(106 examples per second) for both classification and learning, and a high recognition accuracy(up to 95.6%) on the MNIST database.

关键词： RRAM neuromorphic system neuron circuits Hebbian learning pattern classification

Improved thermal property of strained InGaAlAs/AlGaAs quantum wells for 808-nm vertical cavity surface emitting lasers

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Chinese Physics B 2022年第3期31卷 299-305页

作者：赵壮壮荀孟潘冠中孙昀周静涛吴德馨 Institute of Microelectronics Chinese Academy of SciencesBeijing 100029China

The 808-nm vertical cavity surface emitting laser(VCSEL) with strained In_(0.13)Ga_(0.75)Al_(0.12)As/Al_(0.3)Ga_(0.7)As quantum wells is designed and fabricated. Compared with the VCSELs with Al_(0.05)Ga_(0.95)As/Al_(0.3)Ga_(0.7)As quantum wells, the VCSEL with strained In_(0.13)Ga_(0.75)Al_(0.12)As/Al_(0.3)Ga_(0.7)As quantum wells is demonstrated to possess higher power conversion efficiency(PCE) and better temperature stability. The maximum PCE of 43.8% for 10-μm VCSEL is achieved at an ambient temperature of 30°C. The size-dependent thermal characteristics are also analyzed by characterizing the spectral power and output power. It demonstrates that small oxide-aperture VCSELs are advantageous for temperature-stable performance.

关键词： 808-nm VCSEL InGaAlAs/AlGaAs quantum wells thermal property

维普期刊数据库

Reinforcement Learning Based Quantization Strategy Optimal Assignment Algorithm for Mixed Precision

在线全文

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2024年第4期79卷 819-836页

作者： Yuejiao Wang Zhong Ma Chaojie Yang Yu Yang Lu Wei Xi’an microelectronics Technology institute Xi’an710065China

The quantization algorithm compresses the original network by reducing the numerical bit width of the model,which improves the computation speed. Because different layers have different redundancy and sensitivity to databit width. Reducing the data bit width will result in a loss of accuracy. Therefore, it is difficult to determinethe optimal bit width for different parts of the network with guaranteed accuracy. Mixed precision quantizationcan effectively reduce the amount of computation while keeping the model accuracy basically unchanged. In thispaper, a hardware-aware mixed precision quantization strategy optimal assignment algorithm adapted to low bitwidth is proposed, and reinforcement learning is used to automatically predict the mixed precision that meets theconstraints of hardware resources. In the state-space design, the standard deviation of weights is used to measurethe distribution difference of data, the execution speed feedback of simulated neural network accelerator inferenceis used as the environment to limit the action space of the agent, and the accuracy of the quantization model afterretraining is used as the reward function to guide the agent to carry out deep reinforcement learning training. Theexperimental results show that the proposed method obtains a suitable model layer-by-layer quantization strategyunder the condition that the computational resources are satisfied, and themodel accuracy is effectively improved.The proposed method has strong intelligence and certain universality and has strong application potential in thefield of mixed precision quantization and embedded neural network model deployment.

关键词： Mixed precision quantization quantization strategy optimal assignment reinforcement learning neural network model deployment