2016 · 45 citations · 1 references
EngineeringComputer EngineeringComputer ScienceComputer VisionField-programmable Gate ArraysReconfigurable ComputingAccelerator ArchitectureConfigurable Logic BlocksDeep LearningFpga ArchitectureHardware AccelerationReconfigurable AcceleratorsAccelerator ReconfigurationAcceleratorsEnergy-efficient Hardware AccelerationPower-efficient Accelerator DesignReconfigurable Accelerator DesignReconfigurable ArchitecturesReconfigurable Processor Architectures
Convolutional neural networks achieve state‑of‑the‑art accuracy on many computer‑vision tasks, yet convolutions account for over 90 % of their computation, making data movement a major energy bottleneck. The study proposes an accelerator to efficiently support CNNs with large, variable filter sizes and many layers needed for state‑of‑the‑art accuracy. The authors show that AlexNet and VGG16 require millions of weights and billions of MACs, illustrating that the large number of filter weights and channels leads to substantial data movement and high energy consumption. Section 14.5.1.
Deep learning using convolutional neural networks (CNN) gives state-of-the-art accuracy on many computer vision tasks (e.g. object detection, recognition, segmentation). Convolutions account for over 90% of the processing in CNNs for both inference/testing and training, and fully convolutional networks are increasingly being used. To achieve state-of-the-art accuracy requires CNNs with not only a larger number of layers, but also millions of filters weights, and varying shapes (i.e. filter sizes, number of filters, number of channels) as shown in Fig. 14.5.1. For instance, AlexNet [1] uses 2.3 million weights (4.6MB of storage) and requires 666 million MACs per 227×227 image (13kMACs/pixel). VGG16 [2] uses 14.7 million weights (29.4MB of storage) and requires 15.3 billion MACs per 224×224 image (306kMACs/pixel). The large number of filter weights and channels results in substantial data movement, which consumes significant energy.
1