Journal Home
Login
Register
Open Access Journal
Submit a Paper
Propose a Special lssue
Open Access
ARTICLE
Architectural Co-Design and Approximation Strategies for Efficient Deep Neural Network Acceleration in Edge-Oriented Custom Hardware
Issue Vol. 2 No. 01 (2025): Volume 02 Issue 01 --- Section Articles
PDF
Abstract
The exponential growth of deep neural network deployment across edge and embedded platforms has fundamentally transformed the design space of custom hardware accelerators. Unlike cloud-centric computing paradigms, edge-oriented systems impose severe constraints on power consumption, latency, memory bandwidth, silicon area, and reliability, while simultaneously demanding real-time inference accuracy and robustness. This tension has driven a paradigm shift away from monolithic, accuracy-centric neural architectures toward hardware-aware approximation techniques and co-designed accelerator frameworks. This article presents an extensive, theory-driven investigation into deep neural network approximation for custom hardware, situating contemporary design methodologies within a broader historical, architectural, and computational context. Grounded in a comprehensive synthesis of the literature, this work critically examines the evolution of hardware-efficient neural models, compiler-assisted optimization, approximation strategies such as quantization and pruning, and the emergence of edge intelligence frameworks that integrate learning, security, and communication constraints.
The study draws heavily on established survey literature on neural network approximation and hardware acceleration, particularly the foundational analysis of approximation strategies for custom hardware platforms articulated by Wang et al. (2019), while embedding these insights into a wider ecosystem of FPGA, ASIC, and edge-computing research. Through a descriptive and interpretive methodological approach, the article explores how architectural decisions are increasingly informed by workload characteristics, data movement patterns, and deployment environments. The results highlight converging trends toward domain-specific accelerators, compiler-driven optimization pipelines, and lightweight convolutional architectures such as MobileNets, ShuffleNet, and SqueezeNet, which collectively redefine performance-per-watt metrics at the edge. The discussion extends these findings by interrogating unresolved theoretical tensions, including the trade-off between approximation-induced efficiency gains and long-term model robustness, security, and adaptability in federated and decentralized learning scenarios.
By synthesizing architectural, algorithmic, and system-level perspectives, this article contributes a unified conceptual framework for understanding the future trajectory of deep neural network acceleration. It argues that sustainable progress in edge intelligence depends not on isolated innovations but on tightly coupled co-design methodologies that align learning models, hardware substrates, and deployment ecosystems. This work concludes by outlining critical directions for future research, emphasizing the need for cross-layer optimization, trustworthy approximation, and resilient accelerator architectures capable of supporting the next generation of intelligent edge systems.
Keywords
References
1. Li, Y., Yu, Y., Susilo, W., Hong, Z., Guizani, M. Security and privacy for edge intelligence in 5G and beyond networks: challenges and solutions. IEEE Wireless Communications, 28(2), 63–69.
2. Howard, A., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., Adam, H. MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv, 2017.
3. Lin, J., Yu, W., Yang, X., Zhao, P., Zhang, H., Zhao, W. An edge computing based public vehicle system for smart transportation. IEEE Transactions on Vehicular Technology, 69(11), 12635–12651.
4. Wang, E., Davis, J.J., Zhao, R., Ng, H.C., Niu, X., Luk, W., Cheung, P.Y.K., Constantinides, G.A. Deep neural network approximation for custom hardware: Where we’ve been, where we’re going. ACM Computing Surveys, 52, 1–39.
5. Iandola, F., Han, S., Moskewicz, M., Ashraf, K., Dally, W., Keutzer, K. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size. arXiv, 2016.
6. Li, M., Liu, Y., Liu, X., Sun, Q., You, X., Yang, H., Luan, Z., Gan, L., Yang, G., Qian, D. The deep learning compiler: A comprehensive survey. IEEE Transactions on Parallel and Distributed Systems, 32, 708–727.
7. Shawahna, A., Sait, S.M., El-Maleh, A. FPGA-based accelerators of deep learning networks for learning and classification: A review. IEEE Access, 7, 7823–7859.
8. Capra, M., Bussolino, B., Marchisio, A., Shafique, M., Masera, G., Martina, M. An updated survey of efficient hardware architectures for accelerating deep convolutional neural networks. Future Internet, 12, 113.
9. Zhang, X., Zhou, X., Lin, M., Sun, J. ShuffleNet: An extremely efficient convolutional neural network for mobile devices. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 6848–6856.
10. Mittal, S. A survey of FPGA-based accelerators for convolutional neural networks. Neural Computing and Applications, 32, 1109–1139.
11. Li, E., Zhou, Z., Chen, X. Edge intelligence: On-demand deep learning model co-inference with device-edge synergy. Proceedings of the Workshop on Mobile Edge Communications, 31–36.
12. Liu, J., Huang, J., Zhou, Y., Li, X., Ji, S., Xiong, H., Dou, D. From distributed machine learning to federated learning: A survey. Knowledge and Information Systems, 1–33.
13. Krizhevsky, A., Sutskever, I., Hinton, G. ImageNet classification with deep convolutional neural networks. Communications of the ACM, 60, 84–90.
14. Moolchandani, D., Kumar, A., Sarangi, S. Accelerating CNN inference on ASICs: A survey. Journal of Systems Architecture, 113, 101887.
15. Lin, J., Yu, W., Zhang, N., Yang, X., Zhang, H., Zhao, W. A survey on internet of things: architecture, enabling technologies, security and privacy, and applications. IEEE Internet of Things Journal, 4(5), 1125–1142.
16. Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., Chen, L.C. MobileNetV2: Inverted residuals and linear bottlenecks. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 4510–4520.
17. Li, X., Chen, T., Cheng, Q., Ma, S., Ma, J. Smart applications in edge computing: overview on authentication and data security. IEEE Internet of Things Journal, 8(6), 4063–4080.
18. Gholami, A., Kwon, K., Wu, B., Tai, Z., Yue, X., Jin, P., Zhao, S., Keutzer, K. SqueezeNext: Hardware-aware neural network design. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 1719–1719.
19. Liu, L., Chen, X., Lu, Z., Wang, L., Wen, X. Mobile-edge computing framework with data compression for wireless network in energy internet. Tsinghua Science and Technology, 24(3), 271–280.
20. Hass, R., Davies, J. What’s powering artificial intelligence? ARM White Paper, 2019.
21. Li, Y., Chen, C., Liu, N., Huang, H., Zheng, Z., Yan, Q. A blockchain-based decentralized federated learning framework with committee consensus. IEEE Network, 35(1), 234–241.
22. Liu, G., Zhao, H., Fan, F., Liu, G., Xu, Q., Nazir, S. An enhanced intrusion detection model based on improved KNN in WSNs. Sensors, 22(4), 1407.
23. Chollet, F. Xception: Deep learning with depthwise separable convolutions. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1800–1807.
