Loading...
  • 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

Current Issue

    • High Performance Computing
      Stateful logic computation in three-dimensional memristor crossbar array
      HU Yi-hong, MA De-sheng, XU Nuo, WANG Wen-qing, HUANG Cheng-long, FANG Liang
      2023, 45(03): 381-389. doi:
      Abstract ( 183 )   PDF (1084KB) ( 183 )     
      The stateful logic based on memristor is an effective way to break the "von Neumann bottleneck" and realize the Processing In Memory (PIM). However, the current research on stateful logic circuits in memory is mostly based on two-dimensional memristor memory array, and there is a lack of discussion on the implementation of stateful logic in more complex three-dimensional memristor memory array. Compared with the planar two-dimensional array, three-dimensional memristor array has greater storage density and richer device connectivity, which may provide a more flexible matching method for the constructing the stateful logic gates. Therefore, it is necessary to discuss the cascading and achieving process of stateful logic gates in three-dimensional memristor array. In this work, based on the planar stacked 3D memristor array, we study the implementation of complex stateful logic computing process from two aspects: the implementation of basic stateful logic gates and the integrated mapping method supporting cascade. Firstly, the connection relationship of devices in planar stacked 3D memristor arrays is analyzed and summarized. Based on this, the matching requirement of stateful logic gate for two-input Boolean logic is obtained. Secondly, a compound state logic gate is proposed. Logic inputs and logic output shares the same memristor, which can realize a complex logic function in one step (for example, defined as ONOR). It saves the number of steps and devices in complex stateful logic calculation process. Finally, an automatic synthesis mapping method based on complex stateful logic calculation in 3D memristor arrays is presented. Test results on the LGsynth91  benchmark show that, compared with the optimal mapping results in the current two-dimensional array, the proposed comprehensive mapping method based on the three-dimensional memristor array achieves the logic calculation between layers, and saves 41.1% of the used area of the array. After the introduction of the ONOR compound gate, the logical operation steps, the number of memristors and the used area of the array are further reduced by 8.6%, 18.8% and 50.5%, respectively.
      Design of an embedded processor with high reliability
      ZHU Ying, TIAN Zeng, CHEN Ye, JIANG Yi-fei, LI Yan-zhe, LIU Xiao-qiang
      2023, 45(03): 390-397. doi:
      Abstract ( 161 )   PDF (914KB) ( 184 )     
      This paper describes the design of a high-performance embedded processor with high reliability, based on the self-developed Shenwei Instruction Set Architecture (ISA). The processor adopts System-on-Chip (SoC) design technology and AMBA on-chip bus. The third generation of self-developed Shenwei high performance 64-bit CPU core, namely Core3, and multiple standard I/O interface modules including PCIe2.0 and USB2.0 are integrated on the chip. The processor is manufactured using domestic mature process technology, and integrates more than 250 million transistors. The processor can run stably with a core frequency of 800 MHz under a wide environment temperature range (-55℃~125°C). The peak performance of double-precision floating-point number is up to 3.2 GFlops and the maximum power consumption is less than 3.2 W. In order to achieve the design goals of high reliability, low power consumption and high performance, the technical methods and means of the chip structure design, reliability design, low power design and physical implementation are introduced in detail, and the test results of the main technical indexes such as chip frequency, power consumption and yield are given. The processor has been applied in many information devices and has achieved good social benefits.
      Design and experiment of photonic quantum chip for quantum game theory
      ZENG Ru, ZHAN Jun-wei, XUE Shi-chuan, WANG Yi-zhi, WANG Dong-yang, LIU Ying-wen, WU Jun-jie
      2023, 45(03): 398-405. doi:
      Abstract ( 182 )   PDF (923KB) ( 177 )     
      Quantum game theory is an interdisciplinary area combining quantum information and game theory. Theoretical studies demonstrate that quantum games surpass the maximal interest of classical ones, which can be applied to analyze and solve fundamental problems in various fields, such as quantum communication and quantum computing. Aiming at a Bayesian quantum game model with conflictive interests, this paper proposes a programmable photonic quantum chip structure, and completes the quantum game experiment with silicon based optical quantum chip for the first time. By dynamically generating and modulating entangled states on the chip, the experiment demonstrates the advantage of the quantum game over the classical one. This paper shows that photonic quantum chips play a vital role in the research of quantum game theory and provide a promising methodology to study more complicated problems in the field of quantum information.
      Research on factors of heat dissipation of CPU chips in FCBGA package
      CHEN Biao, CHEN Cai, ZHANG Kun, YE Qin
      2023, 45(03): 406-410. doi:
      Abstract ( 266 )   PDF (711KB) ( 278 )     
      Thermal design is a very important part of chip packaging design, which directly affects the temperature and reliability of the chip during operation. The size parameters and physical properties of the packaging materials inside the chip have great influence on the heat dissipation of the chip. The thermal resistance of the chip or the junction temperature can be used to measure the heat dissipation performance. This paper studies the heat dissipation performance of some domestic FCBGA package by numerical simulation (Finite Volume Method), and analyzes the influence of factors such as material size, thermal conductivity of each layer in the CPU package and power density on the CPU temperature and thermal resistance. The research results show that, when the thermal conductivity of TIM1 is lower than  35 W/(m·K), the thermal conductivity and thickness of TIM1 have great influence on the heat dissipation of the CPU; the die area (power density) has great influence on the heat dissipation of the CPU, and the thickness of the die has little effect.
      OpenEmulator: A co-emulation platform based on TSN chip verification
      WANG Zheng, HUANG Rong, WU Mao-wen, SUN Yin-han, SUN Zhi-gang
      2023, 45(03): 411-419. doi:
      Abstract ( 212 )   PDF (1134KB) ( 218 )     
      Hardware emulator is an important means to speed up the verification of Time-Sensitive Networking (TSN) chips. Since TSN chips are far less complex than SoCs (System on Chips), CPU-based hardware emulators can already meet the performance requirements of TSN chip verification. To meet the needs of the TSN chip design, a hardware emulator called OpenEmulator (abbreviated as OE) for the verification of TSN chips is designed and implemented. According to the characteristics of TSN system emulation, a time synchronization mechanism applied to OpenEmulator, called time interlock, is proposed, which realizes precise time synchronization between the physical domain running the real TSN application and the emulation domain running the hardware logic programed by the hardware description language (HDL). At present, OpenEmulator has been applied in the design process of OpenTSN chips. Based on a common PC, OpenEmulator can emulate the initialization of a 6-node TSN network and the subsequent first clock synchronization process in 20 minutes, greatly improving the efficiency of TSN chip emulation verification. Now, OpenEmulator has been open-sourced and integrated into the newly released version of the OpenTSN open source project (version 3.4).
      Circuit optimization of Grover quantum search algorithm
      WU Xi, LI Zhi-qiang, YANG Dong-han
      2023, 45(03): 420-425. doi:
      Abstract ( 123 )   PDF (939KB) ( 184 )     
      Grover algorithm is a quantum search algorithm that can find the target state efficiently. However, with the increase of the amount of searching data, Grovers quantum circuit is faced with the complex gate decomposition problem. In todays NISQ era, resources are very limited, so circuit depth is an important metric. This paper introduces a two-stage quantum search algorithm based on divide-and-conquer, which can run quickly in parallel on a quantum computer. A circuit optimization method is proposed to reduce the number of iterations by using block-level oracle circuit. Combining this method with divide-and-conquer idea, it is defined as 2P-Grover algorithm. The simulation experiment was carried out on the quantum computing framework Cirq.  The experimental results show that, compared with Grover algorithm, the 2P-Grover algorithm can reduce the circuit depth by at least 1.2 times and maintain a high probability of search success.
      Computer Network and Znformation Security
      An optimization method of high density LoRa network
      LI Chao, TU Guo-qing,
      2023, 45(03): 426-433. doi:
      Abstract ( 100 )   PDF (660KB) ( 162 )     
      LoRa technology is widely used in agricultural Internet of Things. When the network scale is large, the standard LoRaWAN protocol can improve the SF to increase the communication distance. However, due to the co-SF and inter-SF interference, LoRaWAN's node density is limited, so it is not suitable for large-scale and high-density deployment scenarios. To meet the requirements of the above deployment scenarios, the multi-hop routing scheme and the SF allocation method are combined to optimize the LoRa network, which improves the node capacity and reliability of the network. Simulation results show that, the proposed method can achieve more than 80% Packet Delivery Rate (PDR) in LoRa network with more than 10 000 nodes, which significantly improves the reliability of LoRa networks in large-scale and high-density scenarios.
      An assured deletion scheme of cloud data based on strongly non-separable cipher
      FU Wei, XIE Zhen-jie, ZHU Ting-ting, REN Zheng-wei
      2023, 45(03): 434-442. doi:
      Abstract ( 118 )   PDF (996KB) ( 150 )     
      Assured deletion of cloud data is a key issue to be solved in the field of cloud storage secu- rity. Existing schemes generally have the drawbacks of over-reliance on key destruction, lack of strong non-separability of ciphertext, excessive encryption and decryption overhead and so on. To solve these problems, by combining AONT conversion with block cipher, a cloud data assured deletion scheme is proposed, which achieves strong non-separability of ciphertext by confusing the original data itself. Theoretical analysis and experimental results show that destroying any piece of cipher data will result in unrecoverable original data in this scheme, thus getting rid of over-reliance on key destruction, which achieves the expected goal of trusted deletion. At the same time, by introducing data block shuffling and reducing the number of cryptographic operations, the ability of anti-ciphertext analysis is improved and the computing overhead is significantly reduced. This scheme has obvious performance advantages compared with existing schemes.
      A verifiable attribute encryption scheme supporting decryption outsourcing in fog computing
      DUAN Ya-hong, WANG Zheng, ZHAO Juan-juan, WANG Long
      2023, 45(03): 443-452. doi:
      Abstract ( 93 )   PDF (850KB) ( 159 )     
      The ciphertext policy-based attribute encryption scheme (CP-ABE) provides secure and fine-grained access control for cloud storage systems. However, due to the large amount of bilinear pairing operation in the encryption and decryption algorithm, it brings a heavy burden to the users. In order to solve the above problems, a verifiable attribute-based encryption scheme supporting outsourced decryption in fog computing is proposed. In the scheme, the linear secret sharing scheme is used to construct the access matrix, which can express various forms of access policies flexibly. Part of the decryption operations is outsourced to fog nodes, so as to reduce the computing burden of the client. To enhance the credibility of outsourced fog nodes, the correctness of the ciphertext accessed by fog nodes is verified through blockchain transactions, and the access behaviors cannot be denied. Through security and experimental analysis, it is proved that the scheme can resist selective plaintext attack and has high operation efficiency.
      Reversible data hiding based on prediction of multiple pairs of asymmetric histograms
      HU Chen-ying, ZHAO Yan
      2023, 45(03): 454-461. doi:
      Abstract ( 78 )   PDF (1094KB) ( 121 )     
      The asymmetric prediction error histogram algorithm can generate two asymmetric histograms that are left by the value of 0 and right by the value of 0. In the embedding process, the two histograms move in opposite directions, so that some modified pixels can be restored to the original pixels, that is, compensation and original reaction will occur. In view of this feature, the algorithm in this paper combines the multi-pair histogram translation algorithm to generate two pairs of asymmetric histograms. Through two rounds of asymmetric histogram embedding, i.e., four layers of data embedding, the probability of compensation restoration of modified pixels is increased. In addition, by studying the change range of peak points and pixel values of the original image after data embedding, all image blocks are classified, and the image blocks with high embedding probability are preferentially selected to reduce unnecessary pixel modification. Compared with other algorithms, this paper makes full use of the compensation and restoration reaction of the asymmetric histogram algorithm, combines the characteristics of the image pixel value distribution to reduce the invalid modification of pixels, and better ensures the quality of the dense image.
      A lattice-based hierarchical certificateless authentication scheme with message recovery for ADS-B
      NONG Qiang, SHAO Meng, ZHANG Bang-bang, LIU Zi-yu,
      2023, 45(03): 462-469. doi:
      Abstract ( 78 )   PDF (791KB) ( 101 )     
      As the key technology of the new generation air traffic control, automatic dependent surveillance-broadcast (ADS-B) has been deployed in most airspace around the world. The existing ADS-B message authentication schemes mainly utilize traditional public key cryptosystem to achieve data security, which are complex for computation and vulnerable to the quantum attack. We apply lattice-based cryptography to ADS-B communication security for the first time, and propose a hierarchical certificateless message authentication scheme supporting message recovery and batch verification simultaneously. The ADS-B airborne equipments are not required to manage certificates, and there is no key escrow problem. The ADS-B messages do not need to be transmitted with the signature, but can be recovered during verification. By utilizing rejection sampling and trapdoor-free technology, the proposed scheme requires just some computationally simple linear operations to realize message authentication. Our scheme is provably secure in the random oracle model under the assumption of the small integer solution (SIS). Experimental results of performance evaluation indicate that this scheme has significant performance improvement in saving computing overhead compared with related works under the same bit security level. It is very suitable for typical aeronautic electronic devices with limited computational resources.
      A real-time facial manipulation video detection model based on ensemble learning dual-stream neural network
      YUAN Ye, HUANG Li-qing, YE Feng, HUANG Tian-qiang, LUO Hai-feng, XU Chao,
      2023, 45(03): 470-477. doi:
      Abstract ( 87 )   PDF (793KB) ( 122 )     
      Malicious face manipulation has a negative impact on social security and stability, and it is a very important issue to accurately detect video images after face tampering. In order to solve the problem of poor real-time performance of video manipulation detection model, this paper proposes a face manipulation video detection model based on ensemble learning dual-stream recurrent neural network, and introduces the voting mechanism in ensemble learning. The model first receives a small number of consecutive sequence frames, extracts spatial features through a convolutional neural network, and introduces central differential convolution to enhance tampering artifacts in the spatial domain. The model then differentiates consecutive sequence frames to enhance tampering artifacts in the temporal domain, while temporal feature extraction is performed through a convolutional neural network. Then, the model splices the dual-stream feature vectors in the spatial domain and the time domain, and performs feature extraction through a recurrent neural network. During the feature extraction process of the recurrent neural network , the frame-by-frame feature information is retained as the input of the subsequent auxiliary frame-level classifier, while the final output of the recurrent neural network is used as the input of the video-level discriminator. Finally, the model introduces the voting mechanism of the integrated model to integrate the outputs of multiple auxiliary frame-level discriminators and video-level discriminators, and introduces a weight hyperparameter γ to balance the importance of the auxiliary frame-level discriminator and video-level discriminator, helping the model to improve detection accuracy. On the FaceForensics++  dataset, the experimental results show that the proposed model  improves the average accuracy by 0.4% and 1.0% compared with mainstream detection model. At the same time, the proposed model can only use fewer consecutive frames for manipulation detection, which improves the real-time performance of the model.
      Graphics and Images
      An image super-resolution reconstruction algorithm based on invertible neural network
      WANG Ping, LI Bin, ZHANG Tong, WANG Jia
      2023, 45(03): 478-488. doi:
      Abstract ( 147 )   PDF (1539KB) ( 179 )     
       In recent years, convolutional neural network has shown good results in single image super-resolution reconstruction task, and has become the most widely used algorithm in this field. However, this algorithm fails to effectively weaken the one-to-many ill-conditioned problem and reduce the solution space range of the reconstructed image, so the effect of this algorithm in improving the image reconstruction quality is becoming more and more limited. At present, it faces the bottleneck problem, and it is difficult to improve the performance greatly. In order to effectively reduce the solution space of reconstructed images and improve the performance of reconstructed images, this paper propose an image super-resolution reconstruction algorithm based on invertible neural network. Through model design, the image degradation and reconstruction process is designed as a reversible transformation process, which effectively constrains the image solution space. The application of invertible convolution structure makes the algorithm obtain the most suitable channel arrangement rules, thus effectively improving the model performance. Experimental results on mainstream data sets show that the proposed algorithm greatly improves the accuracy of image reconstruction compared with the existing SISR algorithm, and achieves the best PSNR and SSIM performance.
       A simulated remote sensing image generation method based on adversarial learning
      MA Zheng, CHU Jun-zheng, WU Peng-fei
      2023, 45(03): 489-494. doi:
      Abstract ( 132 )   PDF (999KB) ( 158 )     
      Remote sensing image data annotation is time-consuming and costly and requires expert knowledge, making it difficult to obtain remote sensing data with labels. Therefore, it is necessary to generate an effective method of remote sensing data with labels. Starting from the cycle-consistent generative adversarial networks for style transfer in the field of computer vision, a simulated remote sensing image conversion method based on deep learning and cycle-consistent generative adversarial networks to generate new dataset is proposed. This method regards the source data and the generated data as the source domain and the target domain, which can be regarded as the style transfer of the simulated remote sensing dataset. The generated dataset can be further used for common tasks of remote sensing images, such as classification, semantic segmentation, and domain adaptation. Experimental results show that this method can effectively generate simulated remote sensing data with style transfer.
      Image semantic segmentation based on feature fusion and attention mechanism
      MA Dong-mei, HUANG Xin-yue, LI Yu
      2023, 45(03): 495-503. doi:
      Abstract ( 148 )   PDF (1320KB) ( 207 )     
      The current high-precision semantic segmentation model requires huge computing resources, so it is difficult to deploy on embedded platforms with limited hardware storage and computing power. Aiming at this issue, an image semantic segmentation model based on feature fusion and attention mechanism is proposed. Firstly, the model based on DeepLabV3+ is optimized and the MobileNetV2 backbone network is lightened using channel pruning. Secondly, the Splittable Triplet Attention (STA) is introduced to the lightweight model to improve the internal dimensional correlation of the feature map. Finally, fine-grained up-sampling modules are added in the decoding part to improve the edge detail information. In the experiments on Pascal VOC 2012 and cityscapes datasets, the parameter number of the proposed algorithm is only 4.15×106, the number of floating-point operations is 10.23 GFLOPs, and the average intersection ratio is 70.98% and 72.26% respectively. The results show that the model achieves a good balance among computing resources, memory consumption and accuracy.
      An improved semantic segmentation algorithm for remote sensing images
      SHE Xiang-yang, MA Yi-jun
      2023, 45(03): 504-511. doi:
      Abstract ( 102 )   PDF (1275KB) ( 163 )     
      Aiming at the problems of edge confusion caused by multiple objects gathering in remote sensing image, unclear segmentation of small scale objects, and insufficient global information acquisition in semantic segmentation process, this paper proposes a semantic segmentation algorithm of remote sensing images based on mixed attention and full-scale skip connection network, called DU-net. In this algorithm, U-net3+ is used as the basic network, and full-scale skip connection network is used as the feature extraction network. The depth supervision in the original model is abandoned, the association between feature and attention mechanism is established, and the process of semantic segmentation is finally realized. The experimental results show that the DU-net algorithm has significant improvement over the classical algorithm under different indexes, and improves the quality of image edge segmentation and the accuracy of the algorithm for small scale target segmentation.
      Artificial Intelligence and Data Mining
      A Chinese sentiment analysis model combining character and word information
      YANG Chun-xia, YAO Si-cheng, SONG Jin-jian,
      2023, 45(03): 512-519. doi:
      Abstract ( 123 )   PDF (700KB) ( 176 )     
      Chinese sentiment analysis models usually only use word granularity information as text representation, which will cause that the model loses the characteristics of word granularity during feature extraction. At the same time, the commonly used word segmentation models are too concise in word segmentation results, which limits the richness of text representation to a certain extent. In this regard, a Chinese sentiment analysis model that combines character granularity features and word granularity features is proposed. The full pattern word segmentation is used to obtain a richer word sequence. After word embedding, the word vector is input into Bi-LSTM to extract the semantic information of the full text. The hidden semantic representation and the corresponding word vector are initially fused to enhance the robustness of word-level information. On the other hand, the word vector is input into multi-window convolution to capture more fine-grained word-level feature information. Finally, the word granularity features are further fused and input into the classifier to obtain the sentiment classification results. The performance test results on two public data sets show that this model improves the classification performance compared with similar models.
      A knowledge graph recommendation model incorporating the influence effect of similar users
      ZHANG Ruo-yi, JIN Liu, MA Hui-fang, WANG Yi-ke, LI Qing-feng
      2023, 45(03): 520-527. doi:
      Abstract ( 103 )   PDF (780KB) ( 164 )     
      Knowledge Graph (KG) owns rich structured information, which can effectively alleviate the sparsity and cold start problem of recommendation model and improve the accuracy and interpretability of recommendation. In recent years, the end-to-end recommendation models incorporating knowledge graph have become a technological trend. This paper proposes a KG recommendation model incorporating the influence effects of similar users. The proposed method expands the interaction between users and items to effectively utilize the knowledge graph. Firstly, a graph neural network neighbor aggregation strategy and an attention mechanism are used to capture two higher-order representations of users and items on the knowledge graph respectively. Then, an influence enhancement layer is designed to capture potential representations of similar users influence effects according to their influence effects. Finally, these three representations are fed together into a multi-layer perceptron to output prediction scores. Experimental results on real datasets show the effectiveness and efficiency of the proposed model.
      A SAIVR epidemic model considering asymptomatic and variant patients
      DENG Yun-feng, LU You-jun, LIANG Yan-jun, ZUO Fei-yu, FU Li
      2023, 45(03): 528-536. doi:
      Abstract ( 120 )   PDF (802KB) ( 140 )     
      Considering the existence of asymptomatic individuals, the existence of mutant individuals, and the direct transformation of susceptible individuals into immune individuals by other means, a new SAIVR infectious disease model is established based on the traditional infectious disease model. According to the propagation rules of the SAIVR model, the propagation dynamics equation of the model is given by using the differential equation theory, and the existence of the disease-free equilibrium point and the endemic equilibrium point of the model is analyzed. The next-generation matrix method is used to compute the basic reproduction number of the model at the equilibrium point of the disease. According to the Routh-Hurwitz criterion, the local asymptotic stability condition of the model at the equilibrium point is obtained, and the global stability of the model is proved by Lyapunov theory. Simulation experiments show that the addition of asymptomatic individuals and mutant individuals can affect the outbreak time, outbreak scale, and death time of infectious diseases. It can effectively control the spread of SAIVR infectious diseases by reducing the transmission rate of infectious diseases in the population and increasing the immunity rate of infected persons and mutants.
      A WSM-TOPSIS group decision-making method based on q-rung orthogonal triangular fuzzy numbers
      WAN Ben-ting, WAN Chun-tao
      2023, 45(03): 537-545. doi:
      Abstract ( 97 )   PDF (492KB) ( 139 )     
      Combining q-Rung orthogonal fuzzy sets and triangular intuitionistic fuzzy numbers, the definition and algorithm of q-Rung orthogonal triangular fuzzy numbers are given. On this basis, the WSM and TOPSIS methods are extended, and a WSM-TOPSIS multi-attribute group decision-making method is proposed. Considering the weights of decision makers and attributes, this method uses WSM to gather the decision matrix proposed by the decision makers for the first time, and calculates the relative closeness of each scheme according to TOPSIS to get the ranking of the pros and cons of the scheme. Finally, the effectiveness and practicability of the method are verified by an example and comparative analysis.