A Dynamically Reconfigurable Accelerator Design Using a Sparse-Winograd Decomposition Algorithm for CNNs
Source abstract
Abstract
Convolutional Neural Networks (CNNs) are widely used in many fields. Due to their high throughput and high level of computing characteristics, however, an increasing number of researchers are focusing on how to improve the computational efficiency, hardware utilization, or flexibility of CNN hardware accelerators. Accordingly, this paper proposes a dynamically reconfigurable accelerator architecture that implements a Sparse-Winograd F(2 2.3 3)-based high-parallelism hardware architecture. This approach not only eliminates the pre-calculation complexity associated with the Winograd algorithm, thereby reducing the difficulty of hardware implementation, but also greatly improves the flexibility of the hardware; as a result, the accelerator can realize the calculation of Conventional Convolution, Grouped Convolution (GCONV) or Depthwise Separable Convolution (DSC) using the same hardware architecture. Our experimental results show that the accelerator achieves a 3x–4.14x speedup compared with the designs that do not use the acceleration algorithm on VGG-16 and MobileNet V1. Moreover, compared with previous designs using the traditional Winograd algorithm, the accelerator design achieves 1.4x–1.8x speedup. At the same time, the efficiency of the multiplier improves by up to 142%.
Source: openalex_inverted_index · Quality: baseline
Research context
OpenAlex topics: CCD and CMOS Imaging Sensors · Advanced Neural Network Applications · Embedded Systems Design Techniques
SciVX discovery tags are terminology-based. OpenAlex topics are source-provided classifications.
References (35)
Show source-provided references
1. ImageNet classification with deep convolutional neural networks
2. Deep residual learning for image recognition
3. Selective search for object recognition10.1007/s11263-013-0620-5 ↗
4. Rich feature hierarchies for accurate object detection and semantic segmentation
5. Learning deconvolution net-work for semantic segmentation
6. Going deeper in facial expression recognition using deep neural networks
7. A deep convolutional neural network based on nested residue number system
8. Fast algorithms for convolutional neural networks
9. Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural networks10.1109/JSSC.2016.2616357 ↗
10. A high energy efficient reconfigurable hybrid neural network processor for deep learning applications10.1109/JSSC.2017.2778281 ↗
11. A 2.9TOPS/W deep convolutional neural network SoC in FD-SOI 28 nm for intelligent embedded systems
12. DNPU: An 8.1TOPS/W reconfigurable CNN-RNN processor for general-purpose deep neural networks
13. Efficient hardware architectures for deep convolutional neural network10.1109/TCSI.2017.2767204 ↗
14. Optimizing the convolution operation to accelerate deep neural networks on FPGA10.1109/TVLSI.2018.2815603 ↗
15. An architecture to accelerate convolution in deep neural networks
16. MobileNets: Efficient convolutional neural networks for mobile vision applications
17. MobileNetV2: Inverted residuals and linear bottlenecks
18. ShuffleNet: An extremely efficient convolutional neural network for mobile devices
19. Xception: Deep learning with depthwise separable convolutions
20. Chen · ACM SIGPLAN Notices · 2014 · 26910.1145/2644865.2541967 ↗
21. DaDianNao: A machine-learning supercomputer
22. Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural net-works10.1109/JSSC.2016.2616357 ↗
23. Han · ACM SIGARCH Computer Architecture News · 2016 · 24310.1145/3007787.3001163 ↗
24. SCNN: An accelerator for compressed-sparse convolutional neural networks
25. Frequency domain acceleration of convolutional neural networks on CPU-FPGA shared memory system
26. Towards a uniform template-based architecture for accelerating 2D And 3D CNNs on FPGA
27. Efficient hardware architectures for deep convolutional neural network10.1109/TCSI.2017.2767204 ↗
28. Evaluating fast algorithms for convolutional neural networks on FPGAs
29. Automatic generation of systemC code from mediator
30. A CNN accelerator on FPGA using depthwise separable convolution10.1109/TCSII.2018.2865896 ↗
31. Automatic optimising CNN with depth-wise separable convolution on FPGA
32. Efficient Sparse-Winograd convolutional neural networks
33. An OpenCL deep learning accelerator on Arria10
34. Optimizing FPGA-based accelerator design for deep convolutional neural works
35. Caffeine: Towards uniformed representation and acceleration for deep convolutional neural networks