Dongrui Fan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Dongrui Fan has a total of 171 co-authors. Here are the top co-authors by number of shared publications:
- Xiaochun Ye
Institute of Computing Technology Chinese Academy of Sciences
51 shared publications - Wenming Li
Institute of Computing Technology Chinese Academy of Sciences
27 shared publications - Mingyu Yan
Chinese Academy of Sciences
23 shared publications - Zhihua Fan
Institute of Computing Technology Chinese Academy of Sciences
14 shared publications - Meng Wu
Institute of Computing Technology, Chinese Academy of Sciences, Beijing, Beijing, China
11 shared publications - Zhimin Zhang
SKLP, Institute of Computing Technology, CAS, Beijing, China
10 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
An ISA-Centric RISC-V Heterogeneous Architecture for Energy Efficient Ray Tracing Pipeline Acceleration
Ray tracing (RT) is increasingly used in interactive graphics and edge rendering, but efficient deployment remains difficult because its two major phases, BVH construction and BVH traversal, have fundamentally different computation patterns, precision requirements, and architectural affinities. Existing designs largely optimize these phases in isolation and therefore do not provide an efficient full-pipeline solution. We present an ISA-centric RISC-V heterogeneous architecture for energy-efficient RT acceleration through phase-aware hardware-software co-design. The design has three components: a full-BVH compression framework that combines exact lossless compression for leaf-triangle blocks with conservative multi-level quantization for AABBs and rays; custom RISC-V ISA extensions for both vectorized BVH-construction kernels and correctness-preserving low-precision traversal; and a phase-aware heterogeneous organization that couples a RVV-enhanced core for construction with a precision-aware traversal coprocessor. For BVH construction, our design achieves speedups of 2.38 × and 1.97 × over XT910 with RVV for SAH-2 and LBVH-2. Compared with Jetson AGX Orin across different BVH widths (BVH2/4/6/8), our design achieves up to 1.59× and 1.77× speedups for SAH and LBVH, while consuming only 20.53%–32.46% of its energy. For BVH traversal, it achieves 1.42 × speedup over AQB8, while consuming only 31.79% of AQB8’s energy. Across different BVH widths, our design achieves up to 2.50× and 2.64× speedups for SAH and LBVH over Orin, while consuming only 27.19%–51.02% of its energy. These results demonstrate that phase-aware ISA-centric design enables efficient, programmable full-pipeline ray tracing.
- 9 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
Open AccessAn ISA-Centric RISC-V Heterogeneous Architecture for Energy Efficient Ray Tracing Pipeline Acceleration
Ray tracing (RT) is increasingly used in interactive graphics and edge rendering, but efficient deployment remains difficult because its two major phases, BVH construction and BVH traversal, have fundamentally different computation patterns, precision requirements, and architectural affinities. Existing designs largely optimize these phases in isolation and therefore do not provide an efficient full-pipeline solution. We present an ISA-centric RISC-V heterogeneous architecture for energy-efficient RT acceleration through phase-aware hardware-software co-design. The design has three components: a full-BVH compression framework that combines exact lossless compression for leaf-triangle blocks with conservative multi-level quantization for AABBs and rays; custom RISC-V ISA extensions for both vectorized BVH-construction kernels and correctness-preserving low-precision traversal; and a phase-aware heterogeneous organization that couples a RVV-enhanced core for construction with a precision-aware traversal coprocessor. For BVH construction, our design achieves speedups of 2.38 × and 1.97 × over XT910 with RVV for SAH-2 and LBVH-2. Compared with Jetson AGX Orin across different BVH widths (BVH2/4/6/8), our design achieves up to 1.59× and 1.77× speedups for SAH and LBVH, while consuming only 20.53%–32.46% of its energy. For BVH traversal, it achieves 1.42 × speedup over AQB8, while consuming only 31.79% of AQB8’s energy. Across different BVH widths, our design achieves up to 2.50× and 2.64× speedups for SAH and LBVH over Orin, while consuming only 27.19%–51.02% of its energy. These results demonstrate that phase-aware ISA-centric design enables efficient, programmable full-pipeline ray tracing.
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 37  (11) , 2327-2341
ESR-HGNN: Eliminating Semantic Redundancy for Efficient Mini-Batch HGNN Inference
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 37  (11) , 2327-2341
ESR-HGNN: Eliminating Semantic Redundancy for Efficient Mini-Batch HGNN Inference
- Open Access
- 20 August 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computer Systems
M^2 -DFU: Multi-Mode Dataflow Architecture for Adaptive and High-Efficiency Data Processing
The impending end of Moore’s Law and Dennard Scaling has prompted a shift in both industry and academia from general-purpose processors (e.g., CPUs and GPGPUs) toward specialized solutions. However, the diverse hardware requirements of edge computing make it challenging for existing specialized architectures to adapt to rapidly evolving algorithms, often prioritizing efficiency at the expense of generality. Our approach leverages the inherent advantages of software-defined dataflow mechanisms and integrates reconfigurable SIMD into the architecture to achieve high efficiency and flexibility in hardware for edge computing, surpassing ASIC architectures. In this work, we propose M ^2 -DFU (Multi-Mode Dataflow Unit), a flexible, highly efficient architecture with configurable multiple execution modes. First, we identify general and specific requirements for various edge applications, including convolutional neural networks (CNNs), digital signal processing (DSP), and typical scientific computing. Next, we design a flexible architecture with high configurability across multiple modes. Efficient and reconfigurable microarchitectural optimizations are implemented to achieve multilevel parallelism and ensure high hardware utilization. Compared to NVIDIA GPUs, M ^2 -DFU achieves an average improvement of up to 6.8 \times in energy efficiency for typical edge computing applications.
- 20 August 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computer Systems
Open AccessM^2 -DFU: Multi-Mode Dataflow Architecture for Adaptive and High-Efficiency Data Processing
The impending end of Moore’s Law and Dennard Scaling has prompted a shift in both industry and academia from general-purpose processors (e.g., CPUs and GPGPUs) toward specialized solutions. However, the diverse hardware requirements of edge computing make it challenging for existing specialized architectures to adapt to rapidly evolving algorithms, often prioritizing efficiency at the expense of generality. Our approach leverages the inherent advantages of software-defined dataflow mechanisms and integrates reconfigurable SIMD into the architecture to achieve high efficiency and flexibility in hardware for edge computing, surpassing ASIC architectures. In this work, we propose M ^2 -DFU (Multi-Mode Dataflow Unit), a flexible, highly efficient architecture with configurable multiple execution modes. First, we identify general and specific requirements for various edge applications, including convolutional neural networks (CNNs), digital signal processing (DSP), and typical scientific computing. Next, we design a flexible architecture with high configurability across multiple modes. Efficient and reconfigurable microarchitectural optimizations are implemented to achieve multilevel parallelism and ensure high hardware utilization. Compared to NVIDIA GPUs, M ^2 -DFU achieves an average improvement of up to 6.8 \times in energy efficiency for typical edge computing applications.
- Open Access
- 24 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
TOSCA: Towards Semi-Centralized Scheduling in Dataflow Architectures via Hardware-Software Co-Design
Dataflow architectures are regarded as promising hardware accelerators in multiple domains due to their efficiency and flexibility. Contemporary applications impose demands on the dynamic scheduling capability of dataflow architectures. Nevertheless, existing approaches face challenges in achieving efficient dynamic scheduling for dataflow architectures: centralized scheduling incurs heavy control overhead due to the limited local agility, whereas decentralized scheduling suffers from constrained planning visibility due to the lack of global awareness. In this paper, we propose TOSCA, a semi-centralized scheduling method for dataflow architectures through hardware-software co-design. Firstly, we introduce a semi-centralized dynamic scheduling execution model, featuring on-demand triggering, global sensing, global-aware planning, and low-latency remapping. Secondly, we present a dataflow microarchitecture designed to support this execution model, synergizing global awareness with local agility through close collaboration of a global-monitoring controller and semi-proactive local PEs. Finally, we propose an adaptive load balancing scheme to efficiently select appropriate PEs as scheduling targets, leveraging the hardware flexibility to enhance parallelism while reducing cost. Experimental results show that in a wide range of representative applications, TOSCA achieves up to 1.52 × performance improvement and 1.40 × energy efficiency improvement over state-of-the-art dataflow architectures.
- 24 July 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
Open AccessTOSCA: Towards Semi-Centralized Scheduling in Dataflow Architectures via Hardware-Software Co-Design
Dataflow architectures are regarded as promising hardware accelerators in multiple domains due to their efficiency and flexibility. Contemporary applications impose demands on the dynamic scheduling capability of dataflow architectures. Nevertheless, existing approaches face challenges in achieving efficient dynamic scheduling for dataflow architectures: centralized scheduling incurs heavy control overhead due to the limited local agility, whereas decentralized scheduling suffers from constrained planning visibility due to the lack of global awareness. In this paper, we propose TOSCA, a semi-centralized scheduling method for dataflow architectures through hardware-software co-design. Firstly, we introduce a semi-centralized dynamic scheduling execution model, featuring on-demand triggering, global sensing, global-aware planning, and low-latency remapping. Secondly, we present a dataflow microarchitecture designed to support this execution model, synergizing global awareness with local agility through close collaboration of a global-monitoring controller and semi-proactive local PEs. Finally, we propose an adaptive load balancing scheme to efficiently select appropriate PEs as scheduling targets, leveraging the hardware flexibility to enhance parallelism while reducing cost. Experimental results show that in a wide range of representative applications, TOSCA achieves up to 1.52 × performance improvement and 1.40 × energy efficiency improvement over state-of-the-art dataflow architectures.
- 12 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. PP  (10459219) , 1-14
FZKP: Alleviating Dataflow Complexity to Exploit Fine-Grained Parallelism for ZKP Acceleration
Zero-knowledge proof (ZKP) is a promising cryptographic protocol, but its practical deployment is hindered by the time-consuming proof generation. The proof generation inherently exhibits high-degree parallelism, yet challenges persist in exploiting fine-grained parallelism due to the dataflow complexity, impeding previous work to achieve optimal acceleration. In this work, we propose FZKP, a ZKP accelerator that utilizes two novel fine-grained dataflows coupled with two forward-flow microarchitectures to alleviate dataflow complexity, efficiently exploiting fine-grained parallelism. The proposed dataflows simplify the dataflow pattern for parallel execution, disclosing fine-grained parallelism at a low cost. The microarchitectures employ a base design to handle large bit-width intermediate results for timely consumption. They then replicate and combine the base design following the proposed dataflow to facilitate parallel execution. When evaluated in 12 nm, FZKP achieves an average speedup of 10.3× and 2.2× over the state-of-the-art GPU-based solution and ZKP accelerator on real-world workloads, respectively.
- 12 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. PP  (10459219) , 1-14
FZKP: Alleviating Dataflow Complexity to Exploit Fine-Grained Parallelism for ZKP Acceleration
Zero-knowledge proof (ZKP) is a promising cryptographic protocol, but its practical deployment is hindered by the time-consuming proof generation. The proof generation inherently exhibits high-degree parallelism, yet challenges persist in exploiting fine-grained parallelism due to the dataflow complexity, impeding previous work to achieve optimal acceleration. In this work, we propose FZKP, a ZKP accelerator that utilizes two novel fine-grained dataflows coupled with two forward-flow microarchitectures to alleviate dataflow complexity, efficiently exploiting fine-grained parallelism. The proposed dataflows simplify the dataflow pattern for parallel execution, disclosing fine-grained parallelism at a low cost. The microarchitectures employ a base design to handle large bit-width intermediate results for timely consumption. They then replicate and combine the base design following the proposed dataflow to facilitate parallel execution. When evaluated in 12 nm, FZKP achieves an average speedup of 10.3× and 2.2× over the state-of-the-art GPU-based solution and ZKP accelerator on real-world workloads, respectively.
- 8 April 2026
- Published by Springer Nature
- p. 230-246
A GCN Accelerator with Unified Architecture
- Open Access
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 239-254
BitRed: Taming Non-Uniform Bit-Level Sparsity with a Programmable RISC-V ISA for DNN Acceleration
The non-uniform and dynamic nature of Bit-Level Sparsity (BLS) poses a critical load-imbalance challenge for parallel hardware accelerators. While the Bit-Interleaving paradigm, represented by state-of-the-art accelerators like Bitlet, shows promise, it is fundamentally constrained by a rigid datapath and severe inter-channel load imbalance. This paper introduces BitRed, an accelerator that embodies a new ''programmable adaptive bit-interleaving'' philosophy. Rather than a monolithic design, BitRed's core is an Adaptive-Sparse Processing Unit (ASPU) that deconstructs the acceleration process into a set of orthogonal RISC-V ISA extensions for pre-processing (cal.pre), adaptive distillation with dynamic load balancing (cal.adis), and PDP-optimal reduction (cal.red). By transforming a rigid hardware problem into a flexible scheduling problem, this ISA-based approach provides a fundamentally more adaptable and extensible solution. Empirical studies on a broad set of benchmarks highlight the following results (normalized to a SCNN baseline): (1) up to 9.4× speedup over the Bitlet, and 5.6× over the latest bit-serial SOTA BitWave; (2) up to 7.6× higher inference efficiency than Bitlet on representative models; (3) 5.072mm2 area and scalable power consumption from 550.43mW (float32) to 495.12mW (16b) and 457.90mW (8b) @28nm TSMC; and (4) high versatility across precisions, and up to 18.9×13.5× higher than NVIDIA A100 and Jetson Orin 32GB, demonstrating significant competitiveness against GPUs.
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 239-254
Open AccessBitRed: Taming Non-Uniform Bit-Level Sparsity with a Programmable RISC-V ISA for DNN Acceleration
The non-uniform and dynamic nature of Bit-Level Sparsity (BLS) poses a critical load-imbalance challenge for parallel hardware accelerators. While the Bit-Interleaving paradigm, represented by state-of-the-art accelerators like Bitlet, shows promise, it is fundamentally constrained by a rigid datapath and severe inter-channel load imbalance. This paper introduces BitRed, an accelerator that embodies a new ''programmable adaptive bit-interleaving'' philosophy. Rather than a monolithic design, BitRed's core is an Adaptive-Sparse Processing Unit (ASPU) that deconstructs the acceleration process into a set of orthogonal RISC-V ISA extensions for pre-processing (cal.pre), adaptive distillation with dynamic load balancing (cal.adis), and PDP-optimal reduction (cal.red). By transforming a rigid hardware problem into a flexible scheduling problem, this ISA-based approach provides a fundamentally more adaptable and extensible solution. Empirical studies on a broad set of benchmarks highlight the following results (normalized to a SCNN baseline): (1) up to 9.4× speedup over the Bitlet, and 5.6× over the latest bit-serial SOTA BitWave; (2) up to 7.6× higher inference efficiency than Bitlet on representative models; (3) 5.072mm2 area and scalable power consumption from 550.43mW (float32) to 495.12mW (16b) and 457.90mW (8b) @28nm TSMC; and (4) high versatility across precisions, and up to 18.9×13.5× higher than NVIDIA A100 and Jetson Orin 32GB, demonstrating significant competitiveness against GPUs.
- 21 January 2026
- Published by Springer Nature in Journal of Computer Science and Technology
- Vol. 41  (2) , 774-790
HARLD: A RISC-V Based Tightly Coupled Heterogeneous Computing Architecture for Low-Density Parity Check Decoding
- 21 January 2026
- Published by Springer Nature in Journal of Computer Science and Technology
- Vol. 41  (2) , 774-790
HARLD: A RISC-V Based Tightly Coupled Heterogeneous Computing Architecture for Low-Density Parity Check Decoding
- Open Access
- 16 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 22  (4) , 1-23
Compressing and Accelerating Sparse CNNs Using Sign-Reserved Toeplitz Filters and Input Activation Density-aware Dataflow
Exploiting the sparsity in convolutional neural networks is crucial to accelerate computing and reduce energy consumption. Unstructured sparsity, benefiting from its flexibility to accommodate arbitrary sparse patterns, generally achieves higher accuracy, but it often introduces irregularity in convolutional operations, which complicates the control logic and undermines the benefits of sparsification. Structured sparsity alleviates these problems but sacrifices its application flexibility, which leads to lower accuracy. In this article, we propose TSCNN, an algorithm-hardware co-design solution that aims to compress and accelerate sparse CNNs while balancing both adaptability to sparsity and computational efficiency. In terms of algorithm, TSCNN adopts pruned filters compressed with sign-reserved Toeplitz matrix format (Tfilters), which systematically enhances the regularity of data reuse and flexibly reduces network parameters by 44%-86% while maintaining accuracy. In terms of hardware, TSCNN accelerator adapts to the structure of Tfilters and utilizes density-aware dataflows to support input activations with large sparsity variation, further optimizing the computational efficiency. Experiments show that TSCNN outperforms a dense CNN accelerator, sparse CNN accelerators SCNN and CSCNN, achieving 5.31× , 2.46× and 1.53× speedup and reducing energy consumption by 80.68%, 69.13% and 53.52%, respectively.
- 16 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 22  (4) , 1-23
Open AccessCompressing and Accelerating Sparse CNNs Using Sign-Reserved Toeplitz Filters and Input Activation Density-aware Dataflow
Exploiting the sparsity in convolutional neural networks is crucial to accelerate computing and reduce energy consumption. Unstructured sparsity, benefiting from its flexibility to accommodate arbitrary sparse patterns, generally achieves higher accuracy, but it often introduces irregularity in convolutional operations, which complicates the control logic and undermines the benefits of sparsification. Structured sparsity alleviates these problems but sacrifices its application flexibility, which leads to lower accuracy. In this article, we propose TSCNN, an algorithm-hardware co-design solution that aims to compress and accelerate sparse CNNs while balancing both adaptability to sparsity and computational efficiency. In terms of algorithm, TSCNN adopts pruned filters compressed with sign-reserved Toeplitz matrix format (Tfilters), which systematically enhances the regularity of data reuse and flexibly reduces network parameters by 44%-86% while maintaining accuracy. In terms of hardware, TSCNN accelerator adapts to the structure of Tfilters and utilizes density-aware dataflows to support input activations with large sparsity variation, further optimizing the computational efficiency. Experiments show that TSCNN outperforms a dense CNN accelerator, sparse CNN accelerators SCNN and CSCNN, achieving 5.31× , 2.46× and 1.53× speedup and reducing energy consumption by 80.68%, 69.13% and 53.52%, respectively.
- Open Access
- 16 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 22  (4) , 1-27
Toward Comprehensive Design Space Exploration on Heterogeneous Multi-core Processors
The theoretical maximum efficiency of heterogeneous multi-core processors (HMPs) is fundamentally determined by their hardware designs. Nonetheless, the achieved efficiency of HMPs is significantly influenced by the strategy employed in resource management (RM). To alleviate the time-consuming nature of comprehensive design space exploration for HMPs, existing work typically follows a route that independently explores HMP designs and RM strategies. However, this route tends to restrict the overall exploration space and efficiency. In this work, we introduce HetDSE, a comprehensive design space exploration framework for HMP design that coordinately explores both HMP designs and RM strategies. We first identify requirements that enable comprehensive exploration, including high-quality design space, fast evaluation, and high-accuracy exploration. To meet these requirements, we propose representative-based design space generation, unified prediction-based evaluation, and iterative refinement exploration methods to achieve comprehensive exploration within a reasonable timeframe. Experimental results demonstrate that HetDSE reduces energy delay square product by 88% compared to the state-of-the-art framework.
- 16 December 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 22  (4) , 1-27
Open AccessToward Comprehensive Design Space Exploration on Heterogeneous Multi-core Processors
The theoretical maximum efficiency of heterogeneous multi-core processors (HMPs) is fundamentally determined by their hardware designs. Nonetheless, the achieved efficiency of HMPs is significantly influenced by the strategy employed in resource management (RM). To alleviate the time-consuming nature of comprehensive design space exploration for HMPs, existing work typically follows a route that independently explores HMP designs and RM strategies. However, this route tends to restrict the overall exploration space and efficiency. In this work, we introduce HetDSE, a comprehensive design space exploration framework for HMP design that coordinately explores both HMP designs and RM strategies. We first identify requirements that enable comprehensive exploration, including high-quality design space, fast evaluation, and high-accuracy exploration. To meet these requirements, we propose representative-based design space generation, unified prediction-based evaluation, and iterative refinement exploration methods to achieve comprehensive exploration within a reasonable timeframe. Experimental results demonstrate that HetDSE reduces energy delay square product by 88% compared to the state-of-the-art framework.