Scholarlink

The proliferation of single-board computers (SBCs) in edge computing necessitates a clear understanding of underlying processor architecture performance for demanding tasks such as real-time malware detection using image recognition. The paper presents a comparative analysis of ARM and RISC-V architectures, embodied by the Raspberry Pi 5 (ARM Cortex-A76) and the Orange Pi RV2 (RISC-V with SiFive U74-class cores), respectively. To classify malware, their ability to efficiently execute a custom, lightweight convolutional neural network (CNN) was assessed. The CNN was trained on the Malimg dataset and then converted to the ONNX model format for cross-platform deployment. The evaluation indicated that the ARM-based Raspberry Pi 5 achieved slightly superior classification accuracy (0.952 vs. 0.944) and F1-scores compared to the RISC-V-based Orange Pi RV2. The ARM platform demonstrated substantially faster inference speeds, processing samples approximately 9.1 times faster than its RISC-V counterpart (0.0074 vs. 0.0672 s per sample). These results highlight the current advantages of ARM’s mature architecture and optimised software ecosystem for compute-intensive edge AI tasks, while underscoring the ongoing development trajectory and potential of the upcoming RISCV ecosystem.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Malware Detection—A Comparative Analysis of RISC-V and ARM Architectures

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Phil Steadman,
  • Paul Jenkins,
  • Rajkumar Singh Rathore,
  • Chaminda Hewage

摘要

The proliferation of single-board computers (SBCs) in edge computing necessitates a clear understanding of underlying processor architecture performance for demanding tasks such as real-time malware detection using image recognition. The paper presents a comparative analysis of ARM and RISC-V architectures, embodied by the Raspberry Pi 5 (ARM Cortex-A76) and the Orange Pi RV2 (RISC-V with SiFive U74-class cores), respectively. To classify malware, their ability to efficiently execute a custom, lightweight convolutional neural network (CNN) was assessed. The CNN was trained on the Malimg dataset and then converted to the ONNX model format for cross-platform deployment. The evaluation indicated that the ARM-based Raspberry Pi 5 achieved slightly superior classification accuracy (0.952 vs. 0.944) and F1-scores compared to the RISC-V-based Orange Pi RV2. The ARM platform demonstrated substantially faster inference speeds, processing samples approximately 9.1 times faster than its RISC-V counterpart (0.0074 vs. 0.0672 s per sample). These results highlight the current advantages of ARM’s mature architecture and optimised software ecosystem for compute-intensive edge AI tasks, while underscoring the ongoing development trajectory and potential of the upcoming RISCV ecosystem.