Scholarlink

This paper proposes a shallow Convolutional Recurrent Neural Network (C-RNN) that directly employs raw signals for classification, implemented on a Field Programmable Gate Array (FPGA) for artificial texture classification. Data was collected using a tactile sensing system based on piezoelectric polymers for eight artificial textures. Preliminary experimental results show that the neural network achieves an accuracy of 96.54%, and the implementation utilizes a total of 208,441 LUTs, 1,247 DSPs, 74,679 flip-flops, and 306 BRAM. The estimated latency is 1,242 clock cycles, corresponding to 6.21 \(\upmu \) s with a 200 MHz clock used in the synthesis.

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

FPGA Implementation of a Convolutional Recurrent Neural Network for Real-Time Sensor Data Processing

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Riccardo Testa,
  • Mohamad Yaacoub,
  • Christian Gianoglio,
  • Maurizio Valle

摘要

This paper proposes a shallow Convolutional Recurrent Neural Network (C-RNN) that directly employs raw signals for classification, implemented on a Field Programmable Gate Array (FPGA) for artificial texture classification. Data was collected using a tactile sensing system based on piezoelectric polymers for eight artificial textures. Preliminary experimental results show that the neural network achieves an accuracy of 96.54%, and the implementation utilizes a total of 208,441 LUTs, 1,247 DSPs, 74,679 flip-flops, and 306 BRAM. The estimated latency is 1,242 clock cycles, corresponding to 6.21 \(\upmu \) s with a 200 MHz clock used in the synthesis.