Recurrent Neural Networks Hardware Implementation on FPGA
Department of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907, USA
arXiv:1511.05552 [cs.NE], (17 Nov 2015)
@article{chang2015recurrent,
title={Recurrent Neural Networks Hardware Implementation on FPGA},
author={Chang, Andre Xian Ming and Martini, Berin and Culurciello, Eugenio},
year={2015},
month={nov},
archivePrefix={"arXiv"},
primaryClass={cs.NE}
}
Recurrent Neural Networks (RNNs) have the ability to retain memory and learn data sequences, and are a recent breakthrough of machine learning. Due to the recurrent nature of RNNs, it is sometimes hard to parallelize all its computations on conventional hardware. CPUs do not currently offer large parallelism, while GPUs offer limited parallelism due to branching in RNN models. In this paper we present a hardware implementation of Long-Short Term Memory (LSTM) recurrent network on the programmable logic Zynq 7020 FPGA from Xilinx. We implemented a RNN with 2 layers and 128 hidden units in hardware and it has been tested using a character level language model. The implementation is more than $21times$ faster than the ARM CPU embedded on the Zynq 7020 FPGA. This work can potentially evolve to a RNN co-processor for future mobile devices.
November 20, 2015 by hgpu
