Low power CNN hardware FPGA implementation

Abstract

A convolution Neural Networks (CNN) goes under the wide umbrella of Deep Neural Networks (DNN) whose applications are widely used. For example, the later are used in robotics and different applications of recognition like speech recognition and facial recognition, also nowadays in autonomous cars. Therefore the aim of implementing the CNN is to be used in real time applications. As a result of that, Graphics processing units (GPUs) are used but their worst disadvantage is it's high power consumption which can't be used in daily used equipments. The target of this paper is to solve the power consumption problem by using Field Programmable Array (FPGA) which has low power consumption, and flexible architecture. The implementation architecture of Alex Network, which consists of three fully connected layers and five convolution layers, on FPGA will depend on two main techniques parallelism of resources, and pipelining inside of some layers.