XCXiaoxin Cui

Xiaoxin Cui

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Xiaoxin Cui has a total of 209 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • A Crosstalk Suppression Scheme Utilizing the Dynamic Shielding and Bit-Stuffing Code for Hexagonal TSV Arrays

    by Chen Wei, Huan Yang, Xiaole Cui, Xing Zhang

    The through silicon via (TSV) arrays play the role of vertical electrical interconnections in the chiplet-based integrated circuits. The hexagonal array achieves a higher TSV density than the rectangular array, but it suffers more severe inter-TSV crosstalk which increases the interconnection delay and deteriorates the signal integrity. Researchers have proposed many crosstalk suppression schemes based on the shielding and crosstalk avoidance code (CAC) techniques, but most of them are applicable only to the rectangular arrays. This paper proposes a crosstalk suppression scheme utilizing the dynamic shielding and bit-stuffing code techniques for hexagonal TSV arrays. Firstly, a novel dynamic shielding scheme is proposed to reduce the maximum crosstalk of the signal TSVs in hexagonal arrays. Then, the bit-stuffing CAC (BS-CAC) is proposed to eliminate the data patterns which cause the high-level crosstalk when these patterns are transmitted through TSVs. The analysis and simulation show that the proposed scheme is able to reduce the inter-TSV crosstalk in hexagonal arrays by 4C-level with the bit overhead of less than 51.5%, and the hardware overhead is less than the existing FTF/FPF-CAC methods.

  • Open Access

    A highly energy-efficient multi-core neuromorphic architecture for training deep spiking neural networks

    by Mingjing Li, Huihui Zhou, Xiaofeng Xu, Zhiwei Zhong, Puli Quan, Xueke Zhu, Yanyu Lin, Wenjie Lin, Xiaosha Li, Dong Wang, et al.

    There is a growing necessity for edge training to adapt to dynamically changing environments. Neuromorphic computing represents a significant pathway for highly efficient intelligent computation in energy-constrained edges, but existing neuromorphic architectures lack the ability of directly training spiking neural networks based on backpropagation. We developed a multi-core neuromorphic architecture with Feedforward-Propagation, Back-Propagation, and Weight-Gradient engines in each core, supporting highly efficient parallel computing at both the engine and core levels, achieving 190% ~ 330% performance of Jetson Orin. It combines various data flows and sparse computation optimization by fully leveraging the sparsity in spiking neural network training, obtaining a high energy efficiency of 1.05TFLOPS/W@ FP16 @ 28 nm, 55 ~ 85% reduction of memory access compared to A100 GPU in the training. Additionally, we deployed the architecture on Field Programmable Gate Arrays, successfully demonstrating 20-core deep spiking network training and 5-worker federated learning. Our study develops the first multi-core neuromorphic architecture supporting direct training of spiking neural network, facilitating neuromorphic computing in edge-learnable applications.


  • A Reconfigurable Built-In Self-Test Scheme for the Evaluation Circuits of Digital SRAM-IMC Architectures

    by Sunrui Zhang, Xiaole Cui, Feng Wei, Xing Zhang

    Digital static random access memory-based in-memory computing (SRAM-IMC) is a promising computation paradigm to break the von-Neumann bottleneck. However, the IMC architectures also bring a series of challenges for testing, because of the circuit structures and operations that do not exist in the conventional memories. One of the challenges is the testing of evaluation circuits in the digital SRAM-IMC architectures, because the primary inputs (PIs) of the evaluation circuits cannot be directly accessed by the testers. Several test approaches such as the conventional logic built-in self-test (LBIST) modules, the indirect and the scan-chain-based test methods are proposed to address this issue. Nevertheless, these solutions suffer from the low test performance or the high area consumption. This work proposes a reconfigurable built-in self-test (BIST) scheme for the evaluation circuits. By reusing the IMC bitcells and operations, the proposed BIST scheme implements the separate pattern generation (PG) and response analysis (RA) processes. Furthermore, the diverse pattern generators, including the Fibonacci linear feedback shift register (LFSR) and weighted LFSR (WLFSR) with adjustable feedback polynomials and the cellular automata (CA), are realized to improve the test efficiency and fault coverage. The evaluation results show that the proposed BIST scheme has better test performance comparing with the indirect and the scan-chain-based test approaches. The proposed BIST scheme has comparable test performance, whereas it has much less area overhead comparing with the conventional LBIST schemes. Additionally, the proposed BIST scheme is testable and repairable.


  • An On-Line BISD Triggering Scheme Based on the CAC-Compatible Error Detection Code for TSV Array

    by Chen Wei, Xiaole Cui, Huan Yang, Xing Zhang

    The through-silicon via (TSV) array plays the significant role of vertical electrical interconnection in the 3-D stacked integrated circuits (SICs). The TSV faults and the inter-TSV crosstalk are two issues that must be addressed for the effective and reliable transmission in TSV arrays. The forbidden transition-free (FTF) and forbidden pattern-free (FPF) crosstalk avoidance code (CAC) can suppress the crosstalk in TSV array, and the built-in self-diagnosis (BISD) and self-repair techniques are able to localize and repair the TSV faults, respectively. The BISD usually consumes a nonnegligible amount of time in the diagnosis mode. So the periodic BISD schemes have a dilemma between the transmission performance and the efficiency of fault detection. The on-demand BISD schemes have been applied to address this issue. This article proposes a new on-line BISD triggering scheme based on the CAC-compatible error detection code (EDC). The proposed scheme checks the data transmitted through the TSV array in the working mode. And the BISD is invoked in time only if the EDC codecs detect errors. Compared with the existing similar schemes, the proposed scheme does not degrade the crosstalk suppression effect of the FTF/FPF-CAC methods, and it achieves a better balance in terms of fault detection ability and hardware overhead.

  • MarchGen: A March Sequence Generation Method for Faults With an Arbitrary Number of Operations in RAMs

    by Sunrui Zhang, Xiaole Cui, Huixian Huang, Feng Wei, Yongliang Chen, Xing Zhang

    The March test is a widely applied test method for memory arrays. Nowadays, the advanced memory technologies keep introducing new types of defects, which bring diverse multioperation fault models in turn. The automatic March sequence generation methods are preferred to avoid the inefficient manual design. Most previous methods generate the March sequences that cover limited types of faults and strongly depend on the specific characteristics of known faults. To address this issue, this work proposes MarchGen, a March sequence generation method for memory faults with an arbitrary number of operations. First, the fault hierarchies, reduced taxonomy, and generic test conditions of 2-composite faults are analyzed. Then, the heuristic March sequence generation workflow is designed accordingly, which takes either fault models or test sequences (TSs) as input. Three optimization strategies are utilized to improve the fault coverage of the generated March sequences and the generation time consumption. The evaluation results show that the proposed method has low dependence on fault types. The generated March sequences reach 100% fault coverage for any simple and 2-composite faults or TSs, and the generation time is roughly linear with the size of the input set.

  • A Joint Data Transmission Scheme With Improved CAC for 3D-SICs Based on Multiple TSV Subarrays

    by Huan Yang, Xiaole Cui, Chen Wei, Juncheng Pu, Zhaohong Lin, Xing Zhang

    Through-silicon via (TSV) is one of the key components enabling vertical electrical interconnects in 3-D stacked integrated circuits (3D-SICs). The high density of TSVs poses substantial crosstalk challenges due to the coupling effects. Traditional crosstalk avoidance codes (CACs) mitigate the crosstalk but bring substantial codec area. The crosstalk suppression capability of CACs is also degraded by the edge effects of TSV array. Although the existing bus partition methods alleviate the drastic codec area increase of CACs, they result in higher bit overhead because the data to be transmitted by each subarray is encoded independently. This work proposes the joint data transmission (JDT) scheme, in which the data to be transmitted by multiple TSV subarrays are encoded jointly. It is proven that the bit overhead of the JDT scheme is less than that of the independent data transmission (IDT) scheme. A systematic design method is proposed to optimize the subarray parameters to minimize the total area of the codecs and TSVs. Furthermore, a new CAC technique named exhaustive reduced adjacent transition CAC (ERAT-CAC) is proposed to improve the crosstalk suppression capability under the edge effects by constraining adjacent signal transitions. The simulation results show that the proposed JDT scheme with the ERAT-CAC has better crosstalk suppression capability under the edge effect compared with fibonacci numeral system (FNS)-forbidden pattern free (FPF)-CAC, FNS-forbidden transition free (FTF)-CAC, and 3D-DPS-CAC, along with smaller area overhead and lower transmission power.


  • A Reconfigurable Digital Compute-in-Memory Architecture With 2:4 Compressed Weight Storage for Highly Sparse Fusion Neural Network

    by Cheng Zhao, Li Lun, Zhenhui Dai, Tingting Zhong, Yi Zhong, Yingying Cui, Xiaoxin Cui

    The fusion of convolutional neural network (CNN) and spiking neural network (SNN) harnesses their complementary strengths in edge video tasks. However, ongoing challenges remain in the low compatibility between two neural networks and the shortage of storage resources. In this brief, we propose a reconfigurable digital compute-in-memory (RDCIM) architecture featuring 2:4 structured weight pruning, tailored for the efficient fusion of CNN and SNN. Specifically, differential frame convolution (DFC) is adopted in place of conventional convolution. Therefore, the RDCIM fusion accelerator can operate in either DFC mode or SNN mode based on the sparsity of activations. First of all, fine-grained 2:4 structured sparsity enables compressed weight storage, thereby doubling the throughput. Secondly, a novel bitwise in-memory Booth multiplier is implemented to support both 4-bit and 8-bit activations, offering flexible computation. Thirdly, the high sparsity inherent in both differential frames and input spikes allows for the concurrent support of three sparse optimization strategies: coarse-grained zero skipping, medium-grained bank gating, and fine-grained precharge gating. Implemented by TSMC 28nm technology, our proposed architecture achieves the peak energy efficiency of 42.02 TOPS/W for DFC and 0.02 pJ/SOP for SNN.


  • A Quantitative Security Ranking Method of PUF Based on the Rademacher Complexity of PUFs

    by Xuexiang Deng, Xiaole Cui, Xing Zhang

    Physical Unclonable Function (PUF) is regarded as one of the promising hardware security primitives. However, the modeling attack poses a real threat to the security of PUFs in recent years. So the security becomes an important characteristic of PUF, in addition to the randomness, uniqueness and reliability. Researchers have proposed some methods to evaluate the security of PUFs. However, the quantitative security ranking method of different PUFs is still an open issue. This work introduces the Rademacher complexity of PUF, abbreviated as the R complexity, to evaluate the security of PUFs. A ranking method of PUF security is proposed based on the R complexity. The proposed method is able to against the noise effect of challenge-response pairs (CRPs). The securities of twelve different types of PUFs with different sizes are ranked by the proposed method. The ranking results are in line with the results from the security improvement practices of these PUFs, which verify the effectiveness of the proposed ranking method.


  • Open Access

    A VC Dimension-Oriented Improvement Method of PUFs for the Anti-Modeling-Attack Capability

    by Yongliang Chen, Xiaole Cui, Sunrui Zhang, Xiaoxin Cui

    The physical unclonable function (PUF) serves as a security primitive of circuits, which is applicable to the embedded systems with lightweight authentication function. However, the modeling attack, which estimates the unknown CRPs by establishing the mathematical model of PUF, is a real threat to the PUF based crypto-systems. Subsequently, the anti-modeling-attack PUF becomes a research hotspot. The systematic design method of secure PUF is still a open issue, although some secure PUF schemes have been proposed based on the repeated trials. This work proposes a security improvement method of PUFs to enhance the anti-modeling-attack capability. The growth function and the Vapnik-Chervonenkis (VC) dimension of PUF are defined as the indicators of PUF security. The proposed method regards the improvement of PUF as an optimization problem, which aims to obtain a PUF scheme with the better security indicators. Guided by the indicators, the proposed method is able to specify the improvement sites of PUF and the techniques to be applied. In addition, three approaches are proposed to inspire the new security improvement techniques. An improved arbiter PUF and an improved array-based PUF are designed as the instances of the results from the proposed method. Both of the improved PUF schemes have the stronger security than the original schemes.


  • FNS-CATF-CAC: An Efficient Crosstalk Avoidance Code to Reduce the Switching Activity in TSV Arrays

    by Chen Wei, Xiaole Cui

    The through silicon via (TSV) arrays play the role of vertical electrical interconnections in the 3-D stacked integrated circuits. However, the coupling crosstalk between the adjacent TSVs increases the interconnection delay and deteriorates the signal integrity in TSV arrays. The crosstalk avoidance code (CAC) techniques based on the Fibonacci numeral system (FNS) or the improved FNS are capable of mitigating the crosstalk in TSV arrays, but the existing schemes are hindered by the hardware overhead, crosstalk suppression ability and switching activity. This article proposes the FNS-based cyclic adjacent transition free CAC with the ouroboros mapping rule for the rectangular and hexagonal TSV arrays. The proposed scheme can reduce the crosstalk even in the presence of the edge effect. Compared with the previous methods, the proposed scheme consumes significantly small hardware overhead in large-scale arrays. And the proposed method can reduce the switching activity on TSVs, thereby alleviating the power consumption in TSV arrays.