近年來卷積式神經網路已在電腦視覺，如自駕車及影像辨識等領域，展現其出色的能力。然而，龐大的運算量、功耗及記憶體需求，仍然是其應用在終端設備時所需面臨的挑戰。針對這個問題，一個可行的方法便是利用二值化神經網路。近期的研究已經證明二值化神經網路能夠在可接受的精確度損失下顯著地降低運算複雜度及記憶體需求。在此篇論文中，我們提出了一種用於二值化神經網路推論的加速器設計。該設計大幅減少所需的運算量並提升了終端設備上的資源利用率。

Convolutional Neural Networks (CNNs) have shown their abilities in computer vision such as self-driving car and image classification applications. However, computation, power consumption, and memory requirement remain challenges for CNNs when applied in the domains of edge devices. To address these challenges, one of the possible solutions is using Binarized Neural Networks (BNNs). Researchers have demonstrated that BNN models dramatically reduce computational complexity and memory requirements with acceptable accuracy loss. This paper presents an accelerator design for BNN inference that minimizes the number of operations and elevates resource utilization on edge devices. We implemented our accelerator on the Xilinx ZCU104 FPGA and evaluated it with TFC, CNV and VGG-13 like BNN networks for MNIST, CIFAR-10 and Tiny ImageNet. Experimental results show that our accelerator can reduce, on average, 99.2\% operations and achieve up to 1974 speedup on an FPGA platform compared to state-of-the-art design.

中文摘要-----------------------------------------i
Abstract----------------------------------------ii
Acknowledgement---------------------------------iii
Contents----------------------------------------v
List of Tables----------------------------------vii
List of Figures---------------------------------viii
1 Introduction----------------------------------1
2 Background------------------------------------3
2.1 Convolutional Neural Networks---------------3
2.2 Binarized Neural Networks-------------------4
3 Proposed Framework----------------------------6
3.1 Multiplexer-based Similarity Operation------6
3.2 Average Redundancy Rate---------------------8
3.3 Accumulation Controller---------------------9
3.4 Pruning Comparator--------------------------11
3.5 Overall Architecture------------------------11
4 Experimental Results--------------------------13
5 Conclusion------------------------------------18

1. J. Bethge, et al., ``MeliusNet: Can binary neural networks achieve mobilenet-level accuracy?,``, in arXiv:2001.05936, 2020
2. Y. Chang, et al., ``A convolutional result sharing approach for binarized neural network inference,`` in Design, Automation Test in Europe Conference Exhibition (DATE), 2020, pp. 780-785.
3. T. Geng, et al., ``LP-BNN: Ultra-low-Latency BNN Inference with Layer Parallelism,`` in International Conference on Application-Specific Systems, Architectures and Processors (ASAP), 2019, pp. 9-16.
4. T. Geng, et al., ``O3BNN-R: An Out-of-Order Architecture for High-Performance and Regularized BNN Inference,`` in IEEE Transactions on Parallel and Distributed Systems, 2020, pp. 199-213.
5. A. G. Howard, et al., ``MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications,`` in arXiv:1704.04861, 2017.
6. M. Kim, P. Smaragdis, ``Bitwise Neural Networks,`` in CoRR, abs/1601.06071, 2016.
7. Q. Liu, et al., ``TCP-Net: Minimizing Operation Counts of Binarized Neural Network Inference,`` in IEEE International Symposium on Circuits and Systems (ISCAS), 2021, pp. 1-5.
8. Q. Liu, et al., ``An Efficient Channel-Aware Sparse Binarized Neural Networks Inference Accelerator,`` in IEEE Transactions on Circuits and Systems II: Express Briefs, 2022, Issue 3, 1637-1641.
9. M. Rastegari, et al., ``XNOR-Net: ImageNet classification using binary convolutional neural networks,`` in Proc. Eur. Conf. Comput. Vis., 2016, pp. 525–542.
10. K. Simonyan, A.Zisserman, ``Very Deep Convolutional Networks for Large-Scale Image Recognition,`` in International Conference on Learning Representations, 2015.
11. W. Tang, et al., ``How to train a compact binary neural network with high accuracy,'' in Proc. 31st AAAI Conf. Artif. Intell., 2017, pp. 2625-2631.
12. Y. Umuroglu, et al., ``Finn: A framework for fast, scalable binarized neural network inference,`` in Proc. ACM/SIGDA Int. Symp.Field-Programmable Gate Arrays, 2017, pp. 65–74.