FPNet: Customized Convolutional Neural Network for FPGA Platforms

Abstract

The Convolutional Neural Network (CNN) has been widely adopted in various applications, which include object detection mobile robot vision, image search engine, etc. And due to the computing-intensive and memory-intensive features of CNN models, specialized hardware accelerators, like Application-Specific-Integrated-Circuit (ASIC) and Field Programmable Gate Arrays (FPGA) have been widely utilized in edge devices. Among all the neural network specialized hardware accelerators, an FPGA accelerator stands out for its flexibility, short time-to-market, and energy efficiency. Previous works about FPGA accelerator designs are mostly changing hardware configuration to fit the CNN model structure. In this paper, we propose an automated neural network architecture search approach utilizing reinforcement learning to design customized convolutional neural network model for specialized FPGA platforms.