YZYouguang Zhang

Youguang Zhang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Youguang Zhang has a total of 75 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Corrections to “A Novel P-bit Unit Based on VGSOT-MTJ for Reconfigurable Ising Machine With Fully Parallel Spin Updating Design”

    by Wentao Huang, Kaili Zhang, Junlin Wang, Yu Liu, Bolin Zhang, Youguang Zhang, Weisheng Zhao, Lang Zeng, Deming Zhang

    In the above article [1], we have found several minor formatting errors. Specifically, there are formatting errors in the display of three parameters in Table I. In addition, there is a black dot appearing at the top of Fig. 2(b), and two black dots at the top of Fig. 3(a). These issues are purely typographical and do not affect the experimental results, analyses, or conclusions presented in the article.TABLE I Parameters ValuesGyromagnetic ratio, $\gamma $ $2.2117 \times 10^{5} \mathrm {~m} /(\mathrm {A} \cdot \mathrm {s})$ PMA coefficient, $\mathrm {K}_{\mathrm {i}}$ $3.2 \times 10^{-4} \mathrm {~J} / \mathrm {m}^{2}$ Exchange bias, $\mathrm {H}_{\mathrm {EX}}$ 20 OeSpin polarization, P 0.58TMR ratio, TMR 100VCMA coefficient, $\beta $ $160 \mathrm {fJ} / \mathrm {V} \cdot \mathrm {m}^{[{21}]}$ Gilbert damping constant, $\alpha $ 0.05Spin hall angle, $\theta _{\text {SH}~}$ 0.25Fig. 2. Fig. 3.

  • A Novel P-bit Unit Based on VGSOT-MTJ for Reconfigurable Ising Machine With Fully Parallel Spin Updating Design

    by Wentao Huang, Kaili Zhang, Junlin Wang, Yu Liu, Bolin Zhang, Youguang Zhang, Weisheng Zhao, Lang Zeng, Deming Zhang

    The spin-based probabilistic bit (p-bit) built on magnetic tunnel junction (MTJ) provides a low power and low overhead solution for building high performance Ising machine (IM). In this work, we propose a novel p-bit device that integrates a voltage-controlled magnetic anisotropy (VCMA) MTJ biased by spin-orbit torque (SOT) voltage. The proposed p-bit achieves an energy consumption of approximately 33fJ/bit per operation. A synchronous Ising network utilizing this device demonstrates rapid convergence to near-ground-state solutions when solving a ${13}\times {26}$ -vertex Max-Cut problem in simulation. Furthermore, a reconfigurable IM architecture, realized through the coupling of MTJ array, enables an all-spin Ising system capable of addressing various combinatorial optimization problems (COPs). The proposed IM system also exhibits functional reconfigurability among multiple invertible Boolean logic gates, achieving a peak accuracy of 95%. These results highlight the potential of the proposed p-bit as a promising building block for high-speed, low-power and universal Ising machines.


  • A Stable FDTD Subgridding Method for Thin PEC Sheet Structure Modeling

    by Langran Deng, Yuhui Wang, Hanhong Liu, Weibo Wu, Xinyue Zhang, Xingqi Zhang, Jian Wang, Youguang Zhang, Zhizhang Chen, Shunchuan Yang

    A finite-difference time-domain (FDTD) subgridding method is developed to efficiently and accurately model thin perfect electric conductor (PEC) sheets. By analyzing two integral loops split by thin sheet structures on the surface of a Yee’s cell, coupling operators at the interface of coarse and subgridding meshes are carefully designed to preserve field continuity and ensure numerical stability. A numerical example is performed to validate the proposed method in terms of stability, accuracy, and efficiency.


  • AM-CIM: Approximate Memory Based Near Sensor Compute-in-Memory Architecture for Keyword Spotting

    by Yuhao Liu, Xiaotao Jia, Guangcai Yuan, Jianyi Yu, Cong Shi, Qi Wei, Youguang Zhang, Weisheng Zhao, Fei Qiao

    Compute-In-Memory (CIM) has emerged as a promising solution to address the von-Neumann bottleneck, making it a key technology for intelligent computing in edge IoT devices, particularly for real-time applications like keyword spotting (KWS). However, traditional CIM architectures face challenges such as high resource consumption, especially in data conversion, which can significantly impact chip area and energy efficiency. To address these challenges, this work proposes a computational CIM architecture utilizing multilevel analog memory, named AM-CIM, tailored for near-sensor (NS) computation of real-time KWS applications. Additionally, approximate memory technology is integrated into the AM-CIM architecture, employing data resilience scheduling for analog memory which contributes to significant reductions in hardware overhead. This integration facilitates a hardware-software co-design approach. To deploy KWS tasks in AM-CIM, a gated recurrent unit (GRU) network, referred to as MAC-GRU, is implemented. By employing Mel-energy as the input feature at the near-sensor end, the system achieves a 93.13% reduction in feature extraction power consumption. Evaluation results based on TSMC 180-nm technology demonstrate that the AM-CIM architecture achieves an accuracy of 88.51% for 10-keyword classification with a power consumption of $546~\mu W$ , while reducing analog memory area by 43.32%.


  • Erratum to “A Reconfigurable Spatial Architecture for Energy-Efficient Inception Neural Networks”

    by Lichuan Luo, Wang Kang, Junzhan Liu, He Zhang, Youguang Zhang, Dijun Liu, Peng Ouyang

    Presents corrections to the paper, (Erratum to “A Reconfigurable Spatial Architecture for Energy-Efficient Inception Neural Networks”).

  • CiTST-AdderNets: Computing in Toggle Spin Torques MRAM for Energy-Efficient AdderNets

    by Lichuan Luo, Erya Deng, Dijun Liu, Zhen Wang, Weiliang Huang, He Zhang, Xiao Liu, Jinyu Bai, Junzhan Liu, Youguang Zhang, et al.

    Recently, Adder Neural Networks (AdderNets) have gained widespread attention as an alternative to traditional Convolutional Neural Networks (CNNs) for deep learning tasks. AdderNets use lightweight addition operations to replace multiplication and accumulation (MAC) operations, but can keep almost the same accuracy compared to other CNNs. Nevertheless, challenges still exist with regards to hardware resources, power consumption, and communication bandwidth, primarily due to the ‘Von-Neumann bottlenecks’. However, computing-in-memory (CIM) architecture based on magnetic random-access memory (MRAM) has great potential for edge DNN implementation. In this paper, we propose a novel CIM paradigm using a novel Toggle-Spin-Torques (TST) driven MRAM for energy-efficient AdderNets (called CiTST_AdderNets). In CiTST_AdderNets, MRAM is driven by the interplay of the field-free spin orbit torque (SOT) effect and the spin transfer torque (STT) effect, which offers a fascinating prospect for energy efficiency and speed. Furthermore, a novel CIM paradigm is proposed to implement the dominating subtraction and sum operations in AdderNets, reducing data transfer and the related energy. Meanwhile, a highly parallel array structure integrating computation and storage is designed to support CiTST_AdderNets. In addition, a mapping strategy is proposed to efficiently map the convolution layer on the array. Fully connected layers can also be efficiently computed. The CiTST-AdderNets macro is designed by using a 65-nm CMOS process. Results show that our CiTST-AdderNets consumes about 1.65 mJ, 9.29 mJ, and 42.46 mJ for running VGG8, ResNet-50, and ResNet-18 respectively at 8-bit fixed-point precision. Compared to state-of-the-art platforms, our macro achieves an energy efficiency improvement of 1.45 x to 66.78 x.


  • Open Access

    Voltage-Controlled Spin-Orbit-Torque-Based Nonvolatile Flip-Flop Designs for Ultra-Low-Power Applications

    by Xiao Liu, Erya Deng, Lichuan Luo, Linjun Jiang, Youguang Zhang, Dijun Liu, Biao Pan, Wang Kang

    Flip-flop (FF) serves as a fundamental unit in various sequential logic circuits and complex digital electronic systems for generating, transforming, and temporarily storing digital signals. Nonvolatility plays a crucial role in FFs by ensuring instant data recovery after unexpected data loss. Nonvolatile flip-flop can quickly recover in a self-powered environment, making it suitable for application environments such as the Internet of Things (IOT). Unfortunately, most existing nonvolatile FFs (NVFFs) suffer from extended delays and high energy consumption during data backup and restore operations. In this paper, we propose two innovative voltage-controlled nonvolatile FFs (VC-FFs), namely VC-DFF (voltage-controlled D-FF) and VC-SRFF (voltage-controlled SR-FF), which address these challenges using voltage-controlled spin-orbit torque (VC-SOT) devices. The proposed designs are evaluated using a 40 nm CMOS process. Simulation results demonstrate that the proposed designs achieve significant improvements in write (recovery) energy consumption, with over 7.2× (1.54×) and 18.7× (2×) enhancements compared to their STT- and SOT-based counterparts, respectively.


  • CiM-BNN:Computing-in-MRAM Architecture for Stochastic Computing Based Bayesian Neural Network

    by Huiyi Gu, Xiaotao Jia, Yuhao Liu, Jianlei Yang, Xueyan Wang, Youguang Zhang, Sorin Dan Cotofana, Weisheng Zhao

    Bayesian neural network (BNN) has gradually attracted researchers' attention with its uncertainty representation and high robustness. However, high computational complexity, large number of sampling operations, and the von-Neumann architecture make a great limitation for the further deployment of BNN on edge devices. In this paper, a new computing-in-MRAM BNN architecture (CiM-BNN) is proposed for stochastic computing (SC)-based BNN to alleviate these problems. In SC domain, neural network parameters are represented in bitstream format. In order to leverage the characteristics of bitstreams, CiM-BNN redesigns the computing-in-memory architecture without complex peripheral circuit requirements and MRAM state flipping. Additionally, real-time Gaussian random number generators are designed using MRAM's stochastic property to further improve energy efficiency. Cadence Virtuoso is used to evaluate the proposed architecture. Simulation results show that energy consumption is reduced more than 93.6% with slight accuracy decrease compared to FPGA implementation with von-Neumann architecture in SC domain.


  • DASM: Data-Streaming-Based Computing in Nonvolatile Memory Architecture for Embedded System

    by Liang Chang, Xin Ma, Zhaohao Wang, Youguang Zhang, Yufei Ding, Weisheng Zhao, Yuan Xie

    Emerging nonvolatile memories (NVMs), including resistive RAM (RRAM), phase-change memory (PCM), and magnetic RAM (MRAM), have opened up new pathways for Computing-In-Memory (CIM). Those NVM technologies can achieve energy-efficient computational operations with only minor modification of the peripheral circuits. Despite many advantages provided by computational NVMs, parallelism is not sufficiently explored in such CIM designs. To break through this limitation on performance gain, we propose a data-streaming design for the NVM-based CIM (e.g., DASM) by leveraging the underlying parallelism in the hardware. DASM benefits from the massive parallelism of data-streaming computing, reduction in data movement of the CIM, and the nonvolatility of memory arrays. Specifically, data streaming operations can be implemented with CIM bitwise operations in both read-out and write-in procedures. In addition, we use the multilevel power gating for the memory array and connections to further boost the performance. Finally, we study a case of inference process for the quantized deep-neural-network-based on the DASM design. DASM architecture achieves 47.8x, 5.1x, 2.1x speedup compared to the NVIDIA Jetson TK1 embedded GPU board, Intel Xeon E5-2640 CPU, the state-of-the-art field-programmable gate array (FPGA) design, with much lower power consumption.


  • Spintronic Processing Unit Within Voltage-Gated Spin Hall Effect MRAMs

    by He Zhang, Wang Kang, Bi Wu, Peng Ouyang, Erya Deng, Youguang Zhang, Weisheng Zhao

    Realization of the unity of processing and storage in the same chip/die has initialized a promising research direction of processing-in-memory (PIM), attempting to address the “memory wall” challenge in typical von-Neumann computing architectures. Nevertheless, there still exists many issues, which requires an efficient implementation of PIM. In this paper, we propose a novel PIM paradigm and architecture to support a spintronic processing unit (SPU), within voltage-gated Spin Hall Effect (VG-SHE) driven MRAMs. Different from previous studies, the data initially stored in the memory units are treated as logic input operands and the logic output results are in-situ stored in the memory units themselves. More importantly, the proposed SPU works just like a typical memory, and all the logic functions are achieved through regular memory-like read/write operations. Besides, a 1-bit full adder based on different array structures (including typical array and crossbar array) is implemented and the possibility of block parallel is discussed. Furthermore, we propose a novel reconfigure architecture, which efficiently translates logical tasks into concrete regular memory-like read/write operations, to support the SPU. Under the 40 nm process technology node, the proposed 1-bit full adder can achieve ~20.00 ns/bit (~8.33 ns/bit) and ~63.8 fJ/bit (~77.35 fJ/bit) based on the typical array (crossbar array) structure.