debughacker / FPGA_Based_CNN

FPGA based acceleration of Convolutional Neural Networks. The project is developed by Verilog for Altera DE5 Net platform.

README

• Open DE5Net_Conv_Accelerator/DE5Net_Conv_Accelerator.qsf with Quartus v16.0.0 on Computer 1

  • Generate Qsys System mem_system
  • Perform Full Compilation
  • Use computer 1 to program the FPGA device installed on HOST Computer (Computer 2) via the USB-Blaster cable. NOTE: This will cause the host computer Kernel to reset and it will automatically reboot. Please save all you work before proceeding with this
  • Reboot the host computer to train the PCIe link with FPGA board.

• Install PCI-Express linux driver on Host PC (Computer 2)

  • Program the FPGA and reboot as mentioned above before performing the installation
  • Install instructions present in README

• Refer to Using the User Application pdf for instructions on data transfer.

This project is a FPGA based implementation of first Convolutional Layer of AlexNet. The accelerator is developed using Verilog.

Contact

Currently Mr. Sachin Kumawat is developing this project. If you have question or need further information, please contact him. Email: [e-mail]

Citing

Our research group is working on acceleration of deep CNNs. If this project helped your research, please kindly cite our latest conference paper:

Mohammad Motamedi, Philipp Gysel, Venkatesh Akella and Soheil Ghiasi, “Design Space Exploration of FPGA-Based Deep Convolutional Neural Network”, IEEE/ACM Asia-South Pacific Design Automation Conference (ASPDAC), January 2016.