ULTRA-LOW-POWER RESNET-50 ON MINIMAL FPGA RESOURCES: A 1.652W, 4-DSP HARDWARE ACCELERATOR FOR EDGE-BASED PNEUMONIA SCREENING
AUTHORS:
Mentor
Jhansi Rani Kaka
Affiliation
Department of Electronics and Communication Engineering, University College of Engineering Kakinada(A), JNTUK, Kakinada, Andhra Pradesh, India
DOI: 
SHARE THIS ARTICLE
CC BY 4.0 License:
This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
DOWNLOAD ARTICLE
REVIEW REPORT
CITE THIS ARTICLE
Abstract
Pneumonia is a progressive pulmonary condition whose radiographic presentation frequently overlaps with other thoracic diseases, presenting major hurdles for automated multi-class diagnosis. Deploying high-performing deep-learning vision models onto resource-constrained edge nodes introduces severe bottlenecks in power, memory, and execution latency. To address this, we evaluate ResNet-50, ConvNeXt V2-Tiny, and Swin Transformer across 17,251 chest X-rays. ResNet-50 achieved 96.19% accuracy (AUC-ROC 0.9929), closely matched by ConvNeXt (~96%) and Swin Transformer (~95%). For edge deployment, the ResNet-50 pipeline was translated into a fixed-point RTL accelerator using Vitis HLS 2025.2 and implemented on a Zynq-7000 (XC7Z010-CLG400-1) SoC via Vivado. Synthesis results demonstrate an ultra-low resource footprint using just 1,658 look-up tables, 2,424 flip-flops, one BRAM block, and four DSP slices, with an on-chip power of 1.652 W and a worst negative slack of 6.874 ns. Cycle-accurate RTL co-simulations confirm real-time throughput and energy efficiency, establishing a robust framework for hardware-accelerated medical diagnostics.
Keywords
Article Information
PUBLICATION DETAILS
Review Type:
Double-Blind Peer Review
License:
CC BY-NC 4.0
Volume/Issue:
Volume 02, Issue 9
Review Rounds:
2
Article Type:
Research Article
Publication Date:
Sep 30 2026
Article Metrics
Article Views
54
PDF Downloads
1
HOW TO CITE
Copy
Ravipati, G. S. S. (2026). Ultra-Low-Power Resnet-50 on Minimal FPGA Resources: A 1.652W, 4-DSP Hardware Accelerator for Edge-Based Pneumonia Screening. International Journal of Science, Strategic Management and Technology, 02(9), 1-9. https://doi.org/10.55041/ijsmt.v2i9.025
Ravipati, Gayathri. "Ultra-Low-Power Resnet-50 on Minimal FPGA Resources: A 1.652W, 4-DSP Hardware Accelerator for Edge-Based Pneumonia Screening." International Journal of Science, Strategic Management and Technology, vol. 02, no. 9, 2026, pp. 1-9. doi:https://doi.org/10.55041/ijsmt.v2i9.025.
Ravipati, Gayathri. "Ultra-Low-Power Resnet-50 on Minimal FPGA Resources: A 1.652W, 4-DSP Hardware Accelerator for Edge-Based Pneumonia Screening." International Journal of Science, Strategic Management and Technology 02, no. 9 (2026): 1-9. https://doi.org/https://doi.org/10.55041/ijsmt.v2i9.025.
References
[1] T. Rahman, A. Khandakar, Y. Qiblawey, A. Tahir, S. Kiranyaz, K. Abulhassan, N. Al-Emadi, M. A.-A. Mohammed, M. U. Khan, “Reliable tuberculosis detection using chest x-ray with deep learning, segmentation and visualization,” IEEE Access, vol. 8, pp. 191586–191601, 2020, doi:10.1109/ACCESS.2020.3031384.
[2] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 770–778, doi:10.1109/CVPR.2016.90.
[3] S. Woo, S. Debnath, R. Hu, X. Chen, Z. Liu, I. S. Kweon, S. Xie, “ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2023, pp. 16133–16142, doi:10.1109/CVPR52729.2023.01548.
[4] Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2021, pp. 10012–10022, doi:10.1109/ICCV48922.2021.00986.
[5] D. S. Kermany, M. Goldbaum, K. Zhang, J. Cai, F. Wang, H. Dong, X. Yuan, M. Cao, C. Ji, X. Zhang, “Identifying medical diagnoses and treatable diseases by image-based deep learning,” Cell, vol. 172, no. 5, pp. 1122–1131, 2018, doi:10.1016/j.cell.2018.02.010.
[6] P. Guddati, S. Dash, R. K. Tripathy, “FPGA implementation of the proposed DCNN model for detection of tuberculosis and pneumonia using CXR images,” IEEE Embedded Syst. Lett., vol. 16, no. 4, pp. 445–448, 2024, doi:10.1109/LES.2024.3370833.
[7] M. Vardhana, R. Pinto, J. Shen, “FPGA-based pulmonary disease detection using lightweight CNN,” IEEE Sensors Lett., vol. 9, no. 10, pp. 1–4, 2025, doi:10.1109/LSENS.2025.3604581.
[8] F. G. Zacchigna, “Methodology for CNN implementation in FPGA-based embedded systems,” IEEE Embedded Syst. Lett., vol. 15, no. 2, pp. 85–88, 2023, doi:10.1109/LES.2023.3243121.
[9] F. Fahim, B. Hawks, C. Herwig, V. Loncar, J. Ngadiuba, N. Tran, T. Aarrestad, J. Duarte, P. Harris, S. Hauck, et al., “hls4ml: An open-source codesign workflow to empower scientific low-power machine learning devices,” arXiv preprint arXiv:2103.05579, 2021.
[10] S. Woo, J. Park, J.-Y. Lee, I. S. Kweon, “CBAM: Convolutional block attention module,” in Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 3–19.
[2] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 770–778, doi:10.1109/CVPR.2016.90.
[3] S. Woo, S. Debnath, R. Hu, X. Chen, Z. Liu, I. S. Kweon, S. Xie, “ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2023, pp. 16133–16142, doi:10.1109/CVPR52729.2023.01548.
[4] Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), 2021, pp. 10012–10022, doi:10.1109/ICCV48922.2021.00986.
[5] D. S. Kermany, M. Goldbaum, K. Zhang, J. Cai, F. Wang, H. Dong, X. Yuan, M. Cao, C. Ji, X. Zhang, “Identifying medical diagnoses and treatable diseases by image-based deep learning,” Cell, vol. 172, no. 5, pp. 1122–1131, 2018, doi:10.1016/j.cell.2018.02.010.
[6] P. Guddati, S. Dash, R. K. Tripathy, “FPGA implementation of the proposed DCNN model for detection of tuberculosis and pneumonia using CXR images,” IEEE Embedded Syst. Lett., vol. 16, no. 4, pp. 445–448, 2024, doi:10.1109/LES.2024.3370833.
[7] M. Vardhana, R. Pinto, J. Shen, “FPGA-based pulmonary disease detection using lightweight CNN,” IEEE Sensors Lett., vol. 9, no. 10, pp. 1–4, 2025, doi:10.1109/LSENS.2025.3604581.
[8] F. G. Zacchigna, “Methodology for CNN implementation in FPGA-based embedded systems,” IEEE Embedded Syst. Lett., vol. 15, no. 2, pp. 85–88, 2023, doi:10.1109/LES.2023.3243121.
[9] F. Fahim, B. Hawks, C. Herwig, V. Loncar, J. Ngadiuba, N. Tran, T. Aarrestad, J. Duarte, P. Harris, S. Hauck, et al., “hls4ml: An open-source codesign workflow to empower scientific low-power machine learning devices,” arXiv preprint arXiv:2103.05579, 2021.
[10] S. Woo, J. Park, J.-Y. Lee, I. S. Kweon, “CBAM: Convolutional block attention module,” in Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 3–19.
Ethics and Compliance
✓ All ethical standards met
This article has undergone plagiarism screening and double-blind peer review. Editorial policies have been followed. Authors retain copyright under CC BY-NC 4.0 license. The research complies with ethical standards and institutional guidelines.
Indexed In
Similar Articles
A Study on Productivity and Employee Motivation at Jersey Dairy PVT. LTD.
(2026)
DOI: 10.55041/ijsmt.v2i8.021
Plant Disease Detection Using a Simple Deep Learning Framework
(2026)
DOI: 10.55041/ijsmt.v2i6.152
Targeting the Gut–Brain–Immune Axis: Emerging Pharmacological Strategies for Neuroinflammatory Disorders
(2026)
DOI: 10.55041/ijsmt.v2i3.268
Previous Article
