An Efficient FPGA-based Depthwise Separable Convolutional Neural Network Accelerator with Hardware Pruning

Abstract

Convolutional neural networks (CNNs) have been widely deployed in computer vision tasks. However, the computation and resource intensive characteristics of CNN bring obstacles to its application on embedded systems. This paper proposes an efficient inference accelerator on FPGA for CNNs with depthwise separable convolutions (DSCs). To improve the accelerator efficiency, we make four contributions: (1) an efficient convolution engine with multiple strategies for exploiting parallelism and a configurable adder tree are designed to support three types of convolution operations; (2) a dedicated architecture combined with input buffers is designed for the bottleneck network structure to reduce data transmission time; (3) a hardware padding scheme to eliminate invalid padding operations is proposed; (4) a hardware-assisted pruning method is developed to support online trade-off between model accuracy and power consumption. Experimental results show that for MobileNetV2 the accelerator achieves 10x and 6x energy efficiency improvement over the CPU and GPU implementation, and 302.3 FPS and 181.8 GOPS performance which is the best among several existing single-engine accelerators on FPGAs. The proposed hardware-assisted pruning method can effectively reduce 59.7% power consumption at the accuracy loss within 5%.