View Code? Open in Web Editor
NEW
FPGA Implementation of the proposed model in master thesis.
C++ 14.53%
Python 3.13%
C 71.86%
Cuda 10.48%
fpga-dcn-kws's Introduction
FPGA Implementation of the proposed model in master thesis using Vivado HLS.
- Naive Convolution (without loop tiling)
- Convolution module using Line Buffer
- GRU (with loop tiling)
- Linear
- Softmax
- Some function from <hls_math> : exp(), tanh()
Target hardware: ZCU 104 (xczu7ev-ffvc1156-2-e)
- Clock: 9.329 ns (Uncertainty 1.25 ns)
- Latency: 580661
- Utilization (%): BRAM_18K (57), DSP48E (22), FF (28), LUT (55)
- AXI4-Stream interface
- Store weights of model on-chip (without DDR)
fpga-dcn-kws's People
Contributors
Stargazers
Watchers