Home Archive
Archive
2025 Volume 62 Issue 6  Published: 2025-06-02
    High Performance Computing System
  • Yulong Zhao , Yanqing Gu , Songtao Tian , Chunzhi Wu , Lingtao Tang , Lufei Zhang , Xiaojun Qin , Xin Liu , Zuoning Chen
    doi: 10.7544/issn1000-1239.202550143

    The number of large-scale language model parameters is growing exponentially, which puts forward higher requirements on the arithmetic density and communication efficiency of the acceleration card, and promotes the rapid development of new architectures, such as single-card multi-core, multi-chip and multi-communication entities. Sunway AI acceleration card adopts four-core group on-chip Ring bus architecture, but in the large model training, the data communication volume is large, and the traditional Ring collection communication method faces the core bottlenecks such as the double limitation of single-core group memory capacity and transmission bandwidth, low collection communication efficiency, and the inability to overlap the communication and computation. In this paper, the optimization framework of SW-IntraCC (Sunway-intra collective communication) is proposed by adopting the concept of software-hardware collaborative design to break through the above limitations through the three-tier storage architecture. First, the three-tier storage architecture is constructed based on on-chip high-speed Ring network, which expands the memory capacity of a single core group by up to four times and increases the host-accelerator card transmission bandwidth by 2.5 times; Second, an intra-chip cross shared communication (CSC) algorithm is designed with interleaved memory access patterns, implementing CSC-AG (CSC-AllGather) and CSC-RS (CSC-ReduceScatter) operators optimized for large model training. Benchmark results demonstrate that CSC achieves 2.15 times higher communication efficiency compared with conventional collective primitives. Finally, a bidirectional operator fusion strategy is proposed to enable communication-computation overlap, yielding a 59% improvement in communication performance after optimization.

  • High Performance Computing System
  • Biao Li , Jie Liu , Qinglin Wang
    doi: 10.7544/issn1000-1239.202550150

    The exact diagonalization (ED) method, a key numerical technique in quantum and condensed matter physics, directly computes the ground state of quantum systems. This study leverages Hamiltonian matrix symmetry, matrix-free methods, hierarchical communication models, and a data-level parallel algorithm optimized for the MT-3000 architecture. We propose a heterogeneous parallel algorithm for large-scale sparse Hamiltonian matrix-vector multiplication on the new generation Tianhe supercomputing system, enabling large-scale exact diagonalization for the one-dimensional Hubbard model. Tests on the Tianhe new generation system show a strong scaling efficiency of 55.27% when scaling from 256 to 8192 processes for a 140-billion-dimension matrix. Weak scaling efficiency remains above 51.25% when scaling from 64 to 13 740 processes for a 730-billion-dimension matrix.

  • High Performance Computing System
  • Rengang Li , Yinan Tang , Zhenhua Guo , Li Wang , Zan Zong , Guangwen Yang
    doi: 10.7544/issn1000-1239.202550120

    With the widespread adoption and development of large-scale artificial intelligence applications, the demand for computing power in artificial intelligence from both industry and academia is increasing. Heterogeneous consistency integrated computing systems, which combine heterogeneous computing technology with cache consistency technology, are gradually becoming an important solution for building intelligent computing centers in the future. However, due to the immaturity of heterogeneous computing and consistency interconnect technologies, it is hard for existing research to model the performance of such systems, making it difficult for researchers to evaluate construction schemes, predict computing performance, and assess system optimization methods at a low cost. We propose HCSim, a performance modeling tool for heterogeneous consistency integrated computing systems, addressing challenges in modeling system topology and inaccuracies in workload modeling within consistency systems. HCSim provides researchers with a flexible, low-cost, and efficient modeling and simulation tool for evaluating interconnect topologies and AI computing tasks. Using HCSim, we model a heterogeneous consistency integrated computing system with thousands of accelerators, and simulate the data-parallel distributed training task of the LLAMA2-13B large language model (LLM) on this system, exploring the impact of variables such as heterogeneous computing power distribution, bandwidth, latency, and task scale on system performance and task execution efficiency. Furthermore, we also design optimization strategies for the communication issues in heterogeneous consistency integrated computing systems and validate the effectiveness of these strategies using HCSim. The simulation results show that HCSim not only meets the performance modeling needs of heterogeneous consistency integrated computing systems, but can also be applied to evaluate and verify optimization strategies for such systems.

  • High Performance Computing System
  • Yan Li , Sile Yang , Chengchun Liu , Linmei Wang , Yaolin Tian , Xinhang Zhang , Yu Zhu , Chunpu Li , Lei Sun , Shengen Yan , Limin Xiao , Weifeng Zhang
    doi: 10.7544/issn1000-1239.202550147

    Large language models with hundreds of billions of parameters are driving rapid technology innovations and business model transformations in artificial intelligence and heterogeneous computing. However, training such models requires prolonged occupation of extensive hardware resources, thus often incurring diverse high-frequency software/hardware failures. These failures not only are challenging to diagnose, but also lead to much longer training time due to unwanted computation waste and slow training convergence. Resilio, an elastic fault-tolerant system for training large language models is proposed, to provide an efficient automated fault recovery mechanism. It is designed to target multiple typical failure scenarios during training processes, such as network interruptions, node crashes, and process failures. Leveraging the characteristics of the parallel model training strategies and underlying hierarchical storage architectures, Resilio implements multi-layer optimizations on checkpoint read/write operations and Just-In-Time (JIT) recovery mechanisms. For models with 100 billion scale parameters, Resilio reduces the end-to-end recovery time under 10 minutes, while reducing the re-started computation after interruptions to the cost of a single training iteration. Upon variations of the computation resources, Resilio can quickly identify the cluster configurations to enable optimal parallel training strategies. Combined with the fault-tolerant scheduling capability, the system ensures adaptive and elastic resource allocations to greatly improve training efficiency and boost GPU utilization across large-scale computing clusters.

  • Intelligent Computing
  • Guopeng Li , Haisheng Tan , Chi Zhang , Hongqiu Ni , Zilong Wang , Xinyue Zhang , Yang Xu , Han Tian , Guoliang Chen
    doi: 10.7544/issn1000-1239.202550128

    Using edge workers to train deep learning recommendation model (DLRM) in edge intelligent computing system brings several benefits, particularly in terms of data privacy protection, low latency and personalization. However, due to the huge size of embedding tables, typical DLRM training frameworks adopt one or more parameter servers to maintain global embedding tables, while leveraging several edge workers to cache part of them. This incurs significant transmission cost for embedding transmissions between workers and parameter servers, which can dominate the training cycle. In this paper, we investigate how to dispatch input embedding samples to appropriate edge workers to minimize the total embedding transmission cost when facing edge-specific challenges such as heterogeneous networks and limited resources. We develop ESD, a novel mechanism that optimizes the dispatching of input embedding samples to edge workers based on expected embedding transmission cost. We propose HybridDis as the dispatch decision method within ESD, which combines a resource-intensive optimal algorithm and a heuristic algorithm to balance decision quality and resource consumption. We implement a prototype of ESD using C++ and Python and compare it with state-of-the-art mechanisms on real-world workloads. Extensive experimental results show that ESD reduces the embedding transmission cost by up to 36.76% and achieves up to 1.74 times speedup in end-to-end DLRM training.

  • Intelligent Computing
  • Yiming Feng , Zhen Qian , Guanghui Li , Chenglong Dai
    doi: 10.7544/issn1000-1239.202550146

    Traditional hierarchical federated learning (HFL) encounters significant challenges in real world due to device heterogeneity, data heterogeneity (e.g., variations in data volume and feature distribution), and communication resource constraints. Device heterogeneity results in inefficient cross-device collaboration during model training, whereas data heterogeneity induces accuracy degradation and diminished generalization capabilities in the global model. To address these limitations while maximizing the utilization of computation, communication, and data resources in the heterogeneous edge networks, we propose an adaptive synergistic method for hierarchical federated learning. This method synergistically integrates model partitioning and client selection under hardware resource constraints, communication bottlenecks, and non-independent and identically distributed (Non-IID) data conditions to accelerate federated learning training while enhancing model accuracy and adaptability across heterogeneous environments. To quantify the influence of local datasets on global model convergence, a data contribution metric is introduced to evaluate the consistency of client contributions. Furthermore, by integrating deep reinforcement learning (DRL) with real-time resource monitoring and data contribution quantification, the DRL agent dynamically optimizes client selection and edge-cloud collaborative model partitioning strategies prior to each training iteration. This adaptive mechanism leverages system resource availability (e.g., bandwidth, device status) and local data contribution scores to derive optimal policies, thereby accelerating training convergence and enhancing global model accuracy. Simulation results demonstrate that the proposed method achieves significant improvements in model accuracy and training efficiency compared with baseline methods, while exhibiting robust adaptability across diverse heterogeneous environment configurations.

  • Compiling and Scheduling Optimization
  • Chuanfu Xu , Haozhong Qiu , Yonggang Che
    doi: 10.7544/issn1000-1239.202550125

    Many high performance computing (HPC) applications such as solving sparse linear systems involve the calculation of sequences of sparse matrix-vector multiplications (SpMV) like Ax, A2x, …, Asx, which is known as the sparse matrix power kernel (MPK). Since MPK calls SpMV using the same sparse matrix, reusing the matrix elements in cache, instead of repeatedly loading them from main memory, can potentially alleviate the memory-constrained problem for SpMV and enhance the performance of MPK. However, reusing the matrix introduces data dependencies between subsequent SpMVs. Prior work mainly either focuses on optimizing an individual SpMV invocation, or introduces significant overheads for cache data reuse in MPK. We propose a cache-aware MPK (Ca-MPK) to optimize MPK via cache data reuse. Based on the dependency graph of a sparse matrix, an architecture-aware recursive partitioning of the graph is designed to obtain subgraphs/submatrices which are fit into cache. Separating subgraphs (i.e., separators) are constructed to decouple data dependencies among subgraphs. Then cache data reuse is achieved by executing sequences of SpMVs on subgraphs with a specified order. Performance evaluation demonstrates that Ca-MPK outperforms the MPK implementations based on the Intel OneMKL library and the state-of-the-art approach, with an average speedup of up to about 1.57 times and 1.40 times, respectively.

  • Compiling and Scheduling Optimization
  • Haotian Wang , Yan Ding , Xianhao He , Guoqing Xiao , Wangdong Yang
    doi: 10.7544/issn1000-1239.202550139

    Sparse matrix-vector multiplication (SpMV) is a core operation in numerical computations, widely used in scientific computing, engineering simulations, and machine learning. The performance optimization of SpMV is mainly constrained by irregular sparse patterns, with traditional optimizations relying on manually designed storage formats, computation strategies, and memory access patterns. Existing tensor compilers such as TACO and TVM use domain-specific languages (DSL) to generate high-performance operators, alleviating developers from the burdensome manual optimization tasks. However, the optimization support for sparse computations remains insufficient, making it challenging to adaptively optimize performance for different sparse patterns. To address these issues, we propose a sparse compiler framework called SparseMode, which generates efficient vectorized code for SpMV computations based on the matrix’s sparse pattern and adaptively adjusts optimization strategies according to the characteristics of the hardware platform. The compiler framework first designs a domain-specific language, SpMV-DSL, which efficiently and concisely expresses sparse matrices and computational operations in SpMV. Then, a sparse pattern-aware method is introduced, which dynamically selects computation strategies based on the matrix storage format and non-zero element distribution defined in SpMV-DSL. Finally, efficient parallel SpMV operator code is generated through sparse pattern analysis and scheduling optimizations, fully utilizing SIMD instructions to enhance performance. Experimental results on different hardware platforms show that SpMV operator code generated by SparseMode achieves up to a 2.44 times speedup compared with the existing TACO and TVM tensor compilers.

  • Compiling and Scheduling Optimization
  • You Fu , Shuhui Jia , Li Chen , Rong Hua , Yunlong Du , Xiran Gao
    doi: 10.7544/issn1000-1239.202550166

    Sequential task flow (STF) represents the access to shared data as dependencies between tasks. The STF runtime system achieves asynchronous parallelism through task construction, dependency analysis, and task dependence graph (TDG) generation, as well as task scheduling. The overhead of these three steps directly affects the performance of parallel programs. The current AceMesh runtime system, with STF at its core, employs a single master core architecture and multiple slave cores for execution on SW39000 processor. However, the discrete memory access performance of SW39000 processor is weak, and the composition of fine-grained tasks increases the discrete memory access, making the composition more likely to become a bottleneck. In this regard, we propose an algorithm that uses multiple auxiliary cores to assist the main core in patterning. First, we analyze the parallelism in the dependency analysis and TDG generation process, and implement a multi-core assisted parallel graph construction algorithm, parallelized fatTDG building algorithm with helpers (PFBH), based on the fat task dependency graph fatTDG on SW39000 processor, and optimize it. Secondly, in response to the problem of main memory resource contention among threads, a method for adjusting the resources of the subordinate cores and parameter selection during parallel graph construction and execution is proposed. Finally, experiments are conducted under five typical applications: compared with a single-core serial graph construction system, the acceleration ratio could achieve up to 1.75 times in fine-grained task scenarios; compared with the OpenACC model on SW39000 processor, AceMesh could achieve up to 2 times acceleration.

  • Compiling and Scheduling Optimization
  • Yuanming Lai , Yalong Li , Hanzhi Hu , Mengyao Xie , Zhe Wang , Chenggang Wu
    doi: 10.7544/issn1000-1239.202550135

    RISC-V, renowned for its open-source nature and modular design, has achieved remarkable success in embedded systems and is progressively expanding into the high-performance computing (HPC) domain. While RISC-V hardware tailored for HPC, such as the Sophon SG2042 multi-core processors, has demonstrated performance comparable to x86/ARM counterparts, its underdeveloped software ecosystem remains a critical barrier to broader adoption. To address this challenge, we develop RVBT, a process-level dynamic binary translator for RISC-V, designed to bridge the software gap by efficiently porting the mature x86 ecosystem to RISC-V platforms, thereby accelerating RISC-V’s integration into HPC applications. Focusing on the pervasive use of SIMD instructions in HPC programs, we tackle the inefficiencies arising from fundamental differences in programming models between x86 SIMD and RISC-V vector (RVV) extensions. Specifically, x86 SIMD hardcodes data types within opcodes, whereas RVV dynamically configures vtype and mask registers, leading to redundant operations during direct translation. To overcome this, we propose three innovative optimizations to achieve: 1) Redundancy elimination via data type locality. By leveraging the locality of data types in adjacent SIMD operations, we statically analyze and remove redundant configurations of vtype (achieving 100% dynamic elimination rates for csrr and vsetvl, and 56.31% for vsetvli) and mask settings (74.66% elimination rate in floating-point benchmarks). 2) Hybrid translation with on-demand synchronization. We decouple scalar and vectorized floating-point operations, translating x86 SIMD scalar double-precision instructions to RISC-V’s floating-point extensions and reserving RVV for vectorized operations. Data synchronization between scalar and vector registers is optimized through def-use analysis, achieving a 67.35% dynamic synchronization reduction in floating-point benchmarks. Experimental results on SPEC CPU 2006 demonstrate significant improvements on the optimized RVBT, achieving 47.39% and 40.06% of native execution efficiency for integer and floating-point benchmarks, respectively, representing speedups of 1.21 times and 8.31 times over the unoptimized version. RVBT vastly outperforms QEMU (18.84% and 4.81% for integer and floating-point), with floating-point efficiency surpassing QEMU by 8.33 times, highlighting its potential for deployment in specific HPC scenarios. Crucially, these optimizations are architecture-agnostic: The methodology of exploiting data type locality, hybrid instruction translation, and adaptive synchronization apply equally to ARM SIMD (e.g., NEON) to RVV translation, offering a universal framework for cross-ISA binary compatibility. This work provides a pivotal technical foundation for breaking the software ecosystem deadlock and advancing RISC-V’s role in HPC.

  • Chip Architecture Design
  • Wenkuo Dong , Chunsuo Yin , Zhimeng Zhang , Pengchao Wang , Jiang Sha , Mengya Wang , Minqi Zhu , Hongwei Liu , Yuhang Liu , Qinfen Hao
    doi: 10.7544/issn1000-1239.202550163

    Wafer-scale computer integrates multiple chiplets through advanced packaging technologies, overcoming traditional chip area limitations to achieve computational power scaling. However, existing domain-specific designs struggle to meet generalized computing requirements. In this study, we propose Yingtian-Lake, which is a wafer-scale general-purpose computer targeting workload characteristics of high-performance computing and intelligent computing scenarios. First, a decoupled computing module-interposer architecture design with standardized I/O interfaces enables multi-modal computing module compatibility. Second, a reconfigurable wafer-scale network employing dynamic topology adaptation technology accommodates diverse traffic patterns. Third, a fault-aware tolerant routing algorithm ensures service continuity during computing unit failures. Experimental results demonstrate that the proposed reconfigurable network achieves second-level topology switching latency. The prototyped 16-module system fabricated with TSMC 28 nm process shows 1.45 times and 1.78 times energy efficiency improvements in high-performance linear algebra computations and deep learning inference tasks respectively, while delivering petaflops-level performance on a single wafer. This breakthrough architecture validates the technical feasibility of universal wafer-scale systems, establishing a scalable hardware foundation for next-generation heterogeneous computing platforms.

  • Chip Architecture Design
  • Ying Xu , Mengdi Wang , Long Cheng , Lian Liu , Shixin Zhao , Lei Zhang , Ying Wang
    doi: 10.7544/issn1000-1239.202550127

    Reinforcement learning with human feedback (RLHF) has been proven effective in aligning large language models (LLMs) with human preferences. The most costly part of RLHF is proximal policy optimization (PPO), which consists of three dependent steps. Different PPO steps in RLHF exhibit different computation modes, simply employing the same parallelization strategy to accelerate all steps that involve multiple model variants, as done in existing frameworks, will lead to poor performance in the PPO generation step due to insufficient utilization of computational resources. Thus, we introduce Pipe-RLHF, a parallelism framework for RLHF fine-tuning, which adaptively employs distinct parallelization strategies for different steps according to the computation mode. Specifically, we first investigate the characteristics of various computation modes to explore their best-fit parallelization approach. And then, we present a novel delayed inter-batch pipeline parallelization approach specifically designed for the PPO generation step, enabling the sufficient utilization of computational resources. Subsequently, based on the proposed inter-batch pipeline parallelization approach, we define a hierarchical parallel plan space for distributed RLHF fine-tuning. Finally, we present optimization algorithms to find the optimal parallelization plan from the defined hierarchical parallel plan space to minimize the overall time consumption. Implementation and evaluation across multiple LLMs demonstrates that the proposed Pipe-RLHF achieves 3.7 times speedup compared with existing methods while achieving near-linear scalability.

  • Chip Architecture Design
  • Qianru Lü , Jinwei Xu , Jingfei Jiang , Dongsheng Li
    doi: 10.7544/issn1000-1239.202550145

    Vision Transformers (ViTs) have demonstrated remarkable success in computer vision tasks, but their complex architecture and computational demands hinder deployment on edge devices. While post-training quantization (PTQ) is widely adopted for model compression, existing PTQ methods exhibit severe performance degradation in 4-bit ultra-low-bitwidth scenarios. This work systematically addresses two fundamental limitations: 1) spatial mismatch between quantization-sensitive layers (e.g., Softmax) and compute-intensive layers (e.g., linear projections), where quantizing Softmax causes 80% accuracy loss despite contributing merely 8% computational load; 2) non-Gaussian activation distributions with hidden Gaussian-like clustering properties (97% values less than three times z-score). We propose DAQ (divide-and-conquer and adaptive quantization), a hardware-friendly PTQ method. DAQ adpots z-score-driven dynamic partitioning algorithm to separate data into normal-range and abnormal-range groups and quantizes the two groups with connected parameter. DAQ further explores hardware accelerated kernel such as tensor core to speed up quantization ViT models. Experimental results demonstrate that DAQ achieves a maximum improvement of 4.37% in ImageNet Top-1 accuracy under 4-bit quantization. In object detection tasks, its average error margin remains below 0.4% compared with the baseline and achieves a maximum improvement of 8.2%, even surpassing the full-precision model by 0.1% in specific cases, thereby realizing near-lossless low-bit-width quantization. Through hardware implementation optimization, DAQ achieves 43%~86% computational acceleration without significantly increasing computational overhead. This approach establishes a synergistic algorithm-hardware co-optimized quantization deployment paradigm for resource-constrained scenarios, effectively balancing model efficiency and precision retention.

  • Chip Architecture Design
  • Hongbo Shi , Zhihua Fan , Wenming Li , Zhiyuan Zhang , Yudong Mu , Xiaochun Ye , Xuejun An
    doi: 10.7544/issn1000-1239.202550160

    Fully homomorphic encryption (FHE), which enables computation on encrypted data without decryption throughout the entire processing flow, offers a promising solution for privacy preservation in cloud computing and other distributed environments. However, the practical deployment of FHE remains significantly constrained by its high computational complexity, poor data locality, and limited parallelism. Among the core operations in FHE, the number theoretic transform (NTT) plays a pivotal role in determining overall system performance. We target the butterfly computation pattern, which is central to the NTT algorithm, and propose a high-efficiency NTT accelerator architecture based on a dataflow computing model. First, we design an RVFHE extension instruction set tailored for NTT butterfly operations, incorporating custom modular multiplication and modular addition/subtraction units to enhance the efficiency of modular arithmetic. Second, we introduce a novel NTT data reordering scheme, combined with a structured butterfly address generation strategy, to reduce the control complexity and access conflicts associated with cross-row and cross-column data exchanges. Finally, we develop a dataflow-driven NTT accelerator architecture that leverages data dependency-triggered execution to enable efficient on-chip scheduling and data reuse, thereby exploiting instruction-level parallelism to the fullest extent. Experimental results demonstrate that, compared with NVIDIA GPU, the proposed architecture achieves up to 8.96 times speedup and 8.53 times improvement in energy efficiency. Furthermore, compared with state-of-the-art dedicated NTT accelerators, our design delivers a 1.37 times performance gain.

  • Chip Architecture Design
  • Ruihao Gao , Shunchen Shi , Xueqi Li , Guangming Tan
    doi: 10.7544/issn1000-1239.202550017

    High-performance and intelligent computing applications require massive data. The transfer and storage of data pose challenges for computer systems. Data compression algorithms reduce storage and transmission costs, making themselves crucial for improving system efficiency. Domain-specific hardware design is an effective way to accelerate data compression algorithms. The emerging data compression software, Zstandard, significantly enhances throughput and compression ratio. Zstandard is based on LZ77 compression algorithm, but it has a larger sliding window that increases on-chip storage overhead. Also, it has complex data dependencies and control flow. These features limit the effect of hardware acceleration. To improve the throughput while achieving a similar compression ratio for data compression accelerator in the context of large sliding windows, we propose a cross-layer optimization approach based on algorithm-architecture co-design to develop a novel data compression acceleration architecture, BeeZip2. First, we introduce the MetaHistory match method into the design of the large sliding window parallel Hash table, offering regular parallelism and addressing control flow data dependency. Then, we propose the shared match PE architecture, distributing the large sliding window across multiple processing units to share on-chip memory and reduce overhead. In addition, the Lazy match strategy and corresponding architecture help to fully leverage the resources for a higher compression ratio. Experimental results show BeeZip2 achieves 13.13 GB/s throughput while maintaining the software compression ratio. Compared with single-core and 36-core CPU software implementations, throughput increases by 29.2 times and 3.35 times, respectively. Compared with the baseline accelerator BeeZip, BeeZip2 achieves a 1.26 times throughput improvement and a 2.02 times throughput-per-area enhancement under the constraint of maintaining a higher compression ratio than its software counterpart.

  • Chip Architecture Design
  • Pengcheng Li , Libo Huang , Gang Chen , Mingche Lai , Lin Deng , Wei Liu , Qianming Yang , Yongwen Wang
    doi: 10.7544/issn1000-1239.202550116

    The rapid development of artificial intelligence (AI) and Internet of things (IoT) technologies has increased demands for the computing energy efficiency, and terminal devices are also facing great challenges in terms of hardware resource overhead. In order to solve the problem of energy efficiency, new designs of approximate computing units with low power have been widely studied. In application scenarios such as digital signal processing and image processing, there are a large number of floating-point operations. These applications consume a lot of hardware resources. However, they have certain fault tolerance characteristics and do not have to perform accurate calculations. In this paper, we propose a 32-bit single-precision reconfigurable approximate floating-point multiplier design method based on multiplication to shift addition (MTA) and asymmetric truncation. Firstly, we utilize an efficient and low-power approximate algorithm that transforms multiplication into shift-addition for partial operands. Secondly, to ensure a balance between accuracy and cost, we propose asymmetric truncation processing for the most significant bits (MSBs) and perform precise calculations on the truncated retained parts. The design achieves high precision and low power. A wide design space is generated by combining the MTA algorithm with different bit widths and varying the number of rows in the partial product array after truncation. This allows for various trade-offs between accuracy and cost. Compared with the exact floating-point multiplier, MTA5T5 has an accuracy loss of mean relative error distance (MRED) of only about 0.32%, a power reduction of 85.80%, and an area reduction of about 79.53%. The accuracy loss of MTA3T3 with lower accuracy is about 1.92%, while the power and area decrease to 90.55% and 85.80%, respectively. Ultimately, we test the application of FIR filtering and image processing, and the results show that the proposed design has significant advantages in terms of accuracy.