Over the past decade, artificial intelligence has reached a stage of rapid development, and deep learning has played a main role in this development. Despite of its strong ability to simulate and predict, deep learning is faced with the problem of large computational complexity. At the hardware level, GPU, ASIC, FPGA are ways to solve the huge amount of computing. This paper will explain the deep learning, FPGA structure and the reason why the use of FPGA to accelerate the deep learning is effective. Also, it will introduce a recursive neural network (RNN) implementation on the FPGA platform.