FAB: FPGA-Accelerated Fully-Pipelined Bottleneck Architecture With Batching for High-Performance MobileNetv2 Inference
IEEE Transactions on Circuits and Systems I: Regular Papers · 2025-11 · 72 (11) · 6615-6628
DOI: 10.1109/tcsi.2025.3573274 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Circuits and Systems I: Regular Papers, 2025-11. Verfasst von Young Chan Kim und 2 weiteren Autor:innen. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 46 Literaturangaben. Gefördert von: Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program, supervised by the IITP.
Abstract (Originalsprache)
Lightweight neural networks (LWNNs) primarily employ the bottleneck block (BB) introduced in MobileNetv2 or similar architectural structures. However, the channel expansion-reduction process in BB imposes substantial activation memory overhead, a challenge that has not been adequately addressed in prior studies on LWNN accelerators incorporating BB. To overcome this limitation, we propose a fully-pipelined bottleneck architecture (FPB) optimized for the efficient hardware deployment of BB. FPB eliminates the need for intermediate off-chip memory access, effectively addressing deployment challenges associated with BB and enabling an end-to-end accelerator architecture. To enhance hardware efficiency, each FPB core utilizes 2-LUT DSP, Fused-ReLU6, and Q-Residual, optimizing computational performance while minimizing resource consumption. Furthermore, we introduce a batching technique that maximizes the benefits of FPB by ensuring high hardware utilization across FPB cores while enabling the concurrent processing of multiple images. To mitigate the off-chip memory access latency inherently incurred by batching, we propose a stem layer latency hiding technique, which effectively prevents performance degradation. We evaluate the performance of our proposed MobileNetv2 accelerator on the VCU118 board, achieving an energy efficiency of 120.7 GOPS/W at a batch size of 4. This represents an improvement of $1.5\times $ to $10.5\times $ over prior work. Depending on the batch size configuration, our FAB accelerator achieves a throughput performance ranging from 204.2 GOPS to 772.7 GOPS, demonstrating its high computational efficiency.
Förderung
- Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education
- MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program
- supervised by the IITP
Zitieren
APA
Kim, Y. C., Kim, N. J., & Kim, H. (2025). FAB: FPGA-Accelerated Fully-Pipelined Bottleneck Architecture With Batching for High-Performance MobileNetv2 Inference. IEEE Transactions on Circuits and Systems I: Regular Papers, 72(11), 6615-6628. https://doi.org/10.1109/tcsi.2025.3573274
Vancouver
Kim YC, Kim NJ, Kim H. FAB: FPGA-Accelerated Fully-Pipelined Bottleneck Architecture With Batching for High-Performance MobileNetv2 Inference. IEEE Transactions on Circuits and Systems I: Regular Papers. 2025;72(11):6615-6628. doi:10.1109/tcsi.2025.3573274
BibTeX
@article{Kim2025,
author = {Kim, Young Chan and Kim, Nam Joon and Kim, Hyun},
title = {FAB: FPGA-Accelerated Fully-Pipelined Bottleneck Architecture With Batching for High-Performance MobileNetv2 Inference},
journal = {IEEE Transactions on Circuits and Systems I: Regular Papers},
year = {2025},
volume = {72},
number = {11},
pages = {6615--6628},
doi = {10.1109/tcsi.2025.3573274}
}
EUVIMED