Optimizing FPGA-based Accelerator Design for Deep Convolutional Neural Networks
The DOI matches its Crossref record; 17 of the 17 references it cites were found in a registry or an index.
Abstract
Convolutional neural network (CNN) has been widely employed for image recognition because it can achieve high accuracy by emulating behavior of optic nerves in living creatures. Recently, rapid growth of modern applications based on deep learning algorithms has further improved research and implementations. Especially, various accelerators for deep CNN have been proposed based on FPGA platform because it has advantages of high performance, reconfigurability, and fast development round, etc. Although current FPGA accelerators have demonstrated better performance over generic processors, the accelerator design space has not been well exploited. One critical problem is that the computation throughput may not well match the memory bandwidth provided an FPGA platform. Consequently, existing approaches cannot achieve best performance due to under-utilization of either logic resource or memory bandwidth. At the same time, the increasing complexity and scalability of deep learning applications aggravate this problem. In order to overcome this problem, we propose an analytical design scheme using the roofline model. For any solution of a CNN design, we quantitatively analyze its computing throughput and required memory bandwidth using various optimization techniques, such as loop tiling and transformation. Then, with the help of rooine model, we can identify the solution with best performance and lowest FPGA resource requirement. As a case study, we implement a CNN accelerator on a VC707 FPGA board and compare it to previous approaches. Our implementation achieves a peak performance of 61.62 GFLOPS under 100MHz working frequency, which outperform previous approaches significantly.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Another option is to use field programable gate arrays (FPGAs), which consume much less energy than GPUs while providing an intermediate computing efficiency between CPUs and GPUs13–17.”
From Hardware implementation of memristor-based artificial neural networks (2024) · cites this paper as [15]
Citations
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 3×
“Figure 2 shows a real-life CNN application, taken from [9].”
BackgroundMethodGradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 1×
“The algorithm receives three 224x224 input images that are from an original 256x256 three-channel RGB image.”
Advances in Neural Information Processing Systems 25
Zhang, Y., Sutton, C. A., Storkey, A., & Zoubin Ghahramani, A. (2012) · Cited by 12,653
3D Convolutional Neural Networks for Human Action Recognition
Ji, S., Xu, W., Yang, M., & Yu, K. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2012.59 · Cited by 6,235
Williams, S., Waterman, A., & Patterson, D. A. (2009) · Communications of the ACM · doi:10.1145/1498765.1498785 · Cited by 2,798
DianNao: a small-footprint high-throughput accelerator for ubiquitous machine-learning
Chen, T., Du, Z., Sun, N., Wang, J., Wu, C., Chen, Y., & Temam, O. (2014) · Proceedings of the 19th international conference on Architectural support for programming languages and operating systems · doi:10.1145/2541940.2541967 · Cited by 1,297
An empirical evaluation of deep architectures on problems with many factors of variation
Larochelle, H., Erhan, D., Courville, A., Bergstra, J., & Bengio, Y. (2007) · Proceedings of the 24th international conference on Machine learning · doi:10.1145/1273496.1273556 · Cited by 992 · cited 1×
“Convolutional neural network (CNN), a well-known deep learning architecture extended from artificial neural network, has been extensively adopted in various applications, which include video surveillance, mobile robot vision, image search engine in data centers, etc [6] [7] [8] [10] [14].”
BackgroundMinimizing Computation in Convolutional Neural Networks
Cong, J., & Xiao, B. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-11179-7_36 · Cited by 372 · cited 1×
“In the feedforward computation perspective, a previous study [5] proved that convolution operations will occupy over 90% of the computation time.”
BackgroundSelf-citationCNP: An FPGA-based processor for Convolutional Networks
Farabet, C., Poulet, C., Han, J. Y., & LeCun, Y. (2009) · 2009 International Conference on Field Programmable Logic and Applications · doi:10.1109/fpl.2009.5272559 · Cited by 367
A dynamically configurable coprocessor for convolutional neural networks
Chakradhar, S., Sankaradas, M., Jakkula, V., & Cadambi, S. (2010) · Proceedings of the 37th annual international symposium on Computer architecture · doi:10.1145/1815961.1815993 · Cited by 343 · cited 8×
“Among these approaches, FPGA based accelerators have attracted more and more attention of researchers because they have advantages of good performance, high energy efficiency, fast development round, and capability of reconfiguration [1] [2] [3] [6] [12] [14].”
BackgroundMethodInfluential
Research works printed but not linked: 1
- D. Aysegul , J. Jonghoon , G. Vinayak , K. Bharadwaj , C. Alfredo , M. Berin , and C. Eugenio . Accelerating deep neural networks on mobile processor with embedded programmable logic . In NIPS 2013 . IEEE, 2013 . D. Aysegul, J. Jonghoon, G. Vinayak, K. Bharadwaj, C. Alfredo, M. Berin, and C. Eugenio. Accelerating deep neural networks on mobile processor with embedded programmable logic. In NIPS 2013. IEEE, 2013.
Cited by
Items 701–800 of 1,002
Fast Convolutional Neural Networks in Low Density FPGAs Using Zero-Skipping and Weight Pruning
Véstias, M. P., Duarte, R. P., de Sousa, J. T., & Neto, H. C. (2019) · Electronics · doi:10.3390/electronics8111321 · Cited by 25
“FPGA implementations of CNNs started with small networks [8,9] and/or considering only convolutional layers [10] and now full CNN implementations [11] and automatic tools for the generation of CNN accelerators [12] are available.”
MethodKim, Y., Kim, H., Yadav, N., Li, S., & Choi, K. K. (2020) · Electronics · doi:10.3390/electronics9030478 · Cited by 25
A High-Accuracy Implementation for Softmax Layer in Deep Neural Networks
Alabassy, B., Safar, M., & El-Kharashi, M. W. (2020) · International Conference on Design & Technology of Integrated Systems in Nanoscale Era (DTIS) · doi:10.1109/dtis48698.2020.9081313 · Cited by 25
“Many hardware accelerator designs were proposed on different platforms from clusters of Graphics Processing Units (GPUs) [4] to more energy-efficient solutions using Field-Programmable Gate Arrays (FPGAs) [5, 6] and Application Specific Integrated Circuits (ASICs) [7, 8].”
Vyasa: A High-Performance Vectorizing Compiler for Tensor Convolutions on the Xilinx AI Engine
Chatarasi, P., Neuendorffer, S., Bayliss, S., Vissers, K., & Sarkar, V. (2020) · IEEE High Performance Extreme Computing Conference · doi:10.1109/hpec43674.2020.9286183 · Cited by 25
“There has been a lot of prior work on optimizing tensor convolutions for a variety of target hardware devices such as CPUs [9], [10], [16], GPUs [9], [16], [17], FPGAs [16], [18]–[21], and Dataflow accelerators [20], [22]–[24].”
BackgroundRoofline-Model-Based Design Space Exploration for Dataflow Techniques of CNN Accelerators
Park, C., Park, S., & Park, C. S. (2020) · IEEE Access · doi:10.1109/access.2020.3025550 · Cited by 25
“The double-buffer structure for overlapping computation and communication from reference [16] is used.”
BackgroundMethodInfluentialA Parametrizable High-Level Synthesis Library for Accelerating Neural Networks on FPGAs
Kalms, L., Rad, P. A., Ali, M. M., Iskander, A., & Göhringer, D. (2021) · Journal of Signal Processing Systems · doi:10.1007/s11265-021-01651-5 · Cited by 25
“While some papers present approaches for cloud applications with sufficient resources [3], others present designs for embedded applications with limited resources [37].”
BackgroundConvolutional Autoencoder-Based Transfer Learning for Multi-Task Image Inferences
Lu, J., Verma, N., & Jha, N. K. (2021) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2021.3068063 · Cited by 25
High-throughput Near-Memory Processing on CNNs with 3D HBM-like Memory
Park, N., Ryu, S., Kung, J., & Kim, J. (2021) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3460971 · Cited by 25
FPGA-based Acceleration for Convolutional Neural Networks on PYNQ-Z2
Huynh, T. V. (2022) · International Journal of Computing and Digital Systems · doi:10.12785/ijcds/110136 · Cited by 25
“With respect to hardware implementations, three hardware platforms can be used as CNN accelerators: graphic processing units (GPU) [5], application-specific integrated circuits (ASICs) [6], and field programmable gate arrays (FPGAs) [7], [8].”
MethodAn efficient EEGNet processor design for portable EEG-Based BCIs
Feng, L., Yang, L., Liu, L., Han, C., Zhang, Y., & Zhu, Z. (2022) · Microelectronics Journal · doi:10.1016/j.mejo.2021.105356 · Cited by 25
F-DNA: Fast Convolution Architecture for Deconvolutional Network Acceleration
Mao, W., Lin, J., & Wang, Z. (2020) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2020.3000519 · Cited by 25
“For example, Zhang et al. [7] proposed a block technique to speed up CNN implementations in parallel.”
BackgroundMethodFast and Energy-Efficient CNN Inference on IoT DevicesPreprint
Motamedi, M., Fong, D. D., & Ghiasi, S. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.07151 · Cited by 25
“There have been different proposals for accelerating CNNs using GPUs ([2], [5]), FPGAs ([8], [4]) and ASICs.”
MethodKung, H. T., McDanel, B., & Zhang, S. Q. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1811.04770 · Cited by 25
“Over the past several years, there has been extensive work on constructing efficient ASIC and FPGA designs for CNNs which generally consider well studied networks such as LeNet5 [34], AlexNet [30], and VGG-16 [55] including [52, 45, 67, 66, 70, 69, 43, 44, 46, 53, 54, 56, 7, 42].”
BackgroundESSA: An energy-Aware bit-Serial streaming deep convolutional neural network accelerator
Hsu, L.-C., Chiu, C., Lin, K.-T., Chou, H.-H., & Pu, Y.-Y. (2020) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2020.101831 · Cited by 24
Hardware accelerator systems for artificial intelligence and machine learning
(2020) · Advances in computers · doi:10.1016/bs.adcom.2020.11.005 · Cited by 24
Accelerating Deep Convolutional Neural Networks Using Number Theoretic Transform
Prasetiyo, Hong, S., Arthanto, Y. F., & Kim, J.-Y. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3214528 · Cited by 24
“The other works [10], [11], [12], [13], [14], [15], [16] focus on specialized hardware architecture and logic design to accelerate the computation itself.”
BackgroundMethodChameleon: Dual Memory Replay for Online Continual Learning on Edge Devices
Aggarwal, S., Binici, K., & Mitra, T. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3347640 · Cited by 24
“Previous works [15], [16] primarily focused on designing effi-cient CNN inference accelerators on FPGAs leveraging various optimization techniques, such as tiling, loop unrolling, and loop transformation.”
BackgroundA Holistic Approach for Optimizing DSP Block Utilization of a CNN implementation on FPGA
Abdelouahab, K., Bourrasset, C., Pelcat, M., Berry, F., Quinton, J., & Sérot, J. (2016) · ACM/IEEE International Conference on Distributed Smart Cameras (ICDSC) · doi:10.1145/2967413.2967430 · Cited by 24
“In [12], an analytical design scheme using the roofline model and loop tiling is used to propose an implementation where the attainable computation roof of the FPGA is reached.”
MethodSong, M., Hu, Y., Xu, Y., Li, C., Chen, H., Yuan, J., & Li, T. (2016) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1145/2967938.2967944 · Cited by 24
“Several recent projects [3, 4, 12, 27, 44] begin to use FPGAs and ASICs as the accelerators to achieve high performance at low energy.”
BackgroundMethodOptimizing Convolutional Neural Networks on the Sunway TaihuLight Supercomputer
Zhao, W., Fu, H., Fang, J., Zheng, W., Gan, L., & Yang, G. (2018) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3177885 · Cited by 24
Parana: A Parallel Neural Architecture Considering Thermal Problem of 3D Stacked Memory
Yin, S., Tang, S., Lin, X., Ouyang, P., Tu, F., Liu, L., Zhao, J., Xu, C., Li, S., Xie, Y., et al. (2018) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2018.2858230 · Cited by 24
“Most prior spatial architecture designs employ isolated on-chip buffers to store input, output, and weight of a neural network, respectively [7], [9], [28], [29], [39], [40].”
InfluentialInstruction Driven Cross-layer CNN Accelerator for Fast Detection on FPGA
Yu, J., Ge, G., Hu, Y., Ning, X., Qiu, J., Guo, K., Wang, Y., & Yang, H. (2018) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3283452 · Cited by 24
“Thus, specific hardware architectures are designed to accelerate CNN on ASIC [3–6] and Field Programmable Gate Array (FPGA) [1, 14, 17, 18, 20, 29].”
MethodBinary Classification of Proteins by a Machine Learning Approach
Perri, D., Simonetti, M., Lombardi, A., Faginas‐Lago, N., & Gervasi, O. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58820-5_41 · Cited by 24
“In addition, CNNs can be easily implemented even on hardware that is not particularly performing, with the advantage of enabling to use the model developed on a plurality of platforms[4] and even on dedicated boards[5].”
BackgroundA Review of FPGA‐Based Custom Computing Architecture for Convolutional Neural Network Inference
Xiyuan, P., Yu, J., Yao, B., Liansheng, L., & Peng, Y. (2021) · Chinese Journal of Electronics · doi:10.1049/cje.2020.11.002 · Cited by 24
PYNQ FPGA Hardware implementation of LeNet-5-Based Traffic Sign Recognition Application
Maraoui, A., Messaoud, S., Bouaafia, S., Ammari, A. C., Khriji, L., & Machhout, M. (2021) · International Multi-Conference on Systems, Signals & Devices (SSD) · doi:10.1109/ssd52085.2021.9429480 · Cited by 24
“Indeed, in many works of CNN accelerators implementation in [6] [7] [8] [9] [10] which have been suggested for different purposes and with various techniques and methodologies.”
BackgroundGonzalez-Carabarin, L., Huijben, I. A. M., Veeling, B., Schmid, A., & van Sloun, R. J. G. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3176809 · Cited by 25
“All these works yield unstructured fine-grained pruning matrices, which leads to inefficiencies in terms of memory access and allocation considering current hardware platforms [28], [29].”
BackgroundN3H-Core: Neuron-designed Neural Network Accelerator via FPGA-based Heterogeneous Computing Cores
Gong, Y. D., Xu, Z., He, Z., Zhang, W., Tu, X., Liang, X., & Jiang, L. (2022) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3490422.3502367 · Cited by 24
Resources and Power Efficient FPGA Accelerators for Real-Time Image Classification
Kyriakos, A., Papatheofanous, E. A., Bezaitis, C., & Reisis, D. I. (2022) · Journal of Imaging · doi:10.3390/jimaging8040114 · Cited by 24
“80 GOP/s of the proposed accelerator; however, considering that both use the same FPGA device, the current work achieves this performance by utilizing only 843 DSPs, compared to the 2240 DSPs of [16] and hence, it results in a significantly higher DSP efficiency of 0.”
BackgroundMethodInfluentialZhang, X., Wei, X. S., Sang, Q., Chen, H., & Xie, Y. (2020) · Electronics · doi:10.3390/electronics9091344 · Cited by 24
“[19] designed a processing engine based on adder tree structure and theoretical roofline model.”
MethodA 7.663-TOPS 8.2-W Energy-efficient FPGA Accelerator for Binary Convolutional Neural NetworksPreprint
Li, Y., Liu, Z., Xu, K., Yu, H., & Ren, F. (2017) · arXiv (Cornell University) · Cited by 24
Enabling Fast and Flexible Distributed Deep Learning with Programmable SwitchesPreprint
Pan, H., Cui, P., li, Z., Jia, R., Zhang, P., Zhang, L., Yang, Y., Wu, J., Dong, J., Cao, Z., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2205.05243 · Cited by 24
“Indeed, training a sparseDLmodel is a time-consuming task, and the distributed sparse DL training, which leverages a cluster of nodes to perform training tasks cooperatively, emerges and becomes a practice [49, 58].”
BackgroundA Survey on Real-Time Object Detection on FPGAs
Hozhabr, S. H., & Giorgi, R. (2025) · IEEE Access · doi:10.1109/access.2025.3544515 · Cited by 25
“Therefore, loop modification techniques are frequently employed in hardware accelerators to optimize performance [203], [257], [258], [259]. when designing real-time object detection systems on FPGAs, techniques such as loop reordering, loop unrolling, and loop pipelining can be employed to…”
Gao, L. H., Luo, Z., & Wang, L. (2025) · Information · doi:10.3390/info16100914 · Cited by 23
Role of Reconfigurable Computing in speeding up Machine Learning Algorithms
Monir, N. I., Akter, F. Y., & Sayed, S. R. K. (2025) · SCCTS Transactions on Reconfigurable Computing · doi:10.31838/rcc/02.02.02 · Cited by 23
Efficient fast convolution architectures for convolutional neural network
Xu, W., Wang, Z., You, X., & Zhang, C. (2017) · IEEE International Conference on ASIC (ASICON) · doi:10.1109/asicon.2017.8252623 · Cited by 23
“Zhang [6] proposed a method aiming at acceleration across different layers with the help of roofline model.”
MethodSelf-citationKosuge, A., & Oshima, T. (2019) · IEEE International Conference on Graph Computing (GC) · doi:10.1109/gc46384.2019.00018 · Cited by 23
“In addition, owing to recent intensive research activities, such as development of quantization techniques [4,5] and energy efficient hardware (HW) techniques for low-power applications [6-8], the CNN incurs little computation time even on low-power hardware.”
BackgroundAccelerating DNNs from local to virtualized FPGA in the Cloud: A survey of trends
Wu, C., Fresse, V., Suffran, B., & Konik, H. (2021) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2021.102257 · Cited by 23
Wang, H., Chen, H., Wang, B., Jin, Y., Li, G., & Kan, Y. (2022) · Applied Energy · doi:10.1016/j.apenergy.2022.119203 · Cited by 23
EDCompress: Energy-Aware Model Compression for Dataflows
Wang, Z., Luo, T., Goh, R. S. M., & Zhou, J. T. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3172941 · Cited by 23
“Dataflow Applied by Dataflow Applied by X : Y [4] [31] FX : FY [5] X : FX [6] [32] [33] CI : CO [7] [34] [35] [36] [37] [38]”
MethodAn efficient lightweight CNN acceleration architecture for edge computing based-on FPGA
Wu, R., Liu, B., Fu, P., & Chen, H. (2022) · Applied Intelligence · doi:10.1007/s10489-022-04251-3 · Cited by 23
“In the early research on CNN acceleration, the roofline model was proposed as the basic theory to solve the balance problem between through-put and bandwidth [30].”
MethodOptimizing CNNs on Multicores for Scalability, Performance and Goodput
Rajbhandari, S., He, Y., Ruwase, O., Carbin, M., & Chilimbi, T. M. (2017) · International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) · doi:10.1145/3037697.3037745 · Cited by 23
“GPUs [27], FPGAs [23, 45, 55] and ASICs [9, 10] have being used to accelerate CNN inference, which is a subset (i.e., forward propagation phase only) of CNN training.”
MethodEfficient Deep Neural Network Serving: Fast and Furious
Yan, F., He, Y., Ruwase, O., & Smirni, E. (2018) · IEEE Transactions on Network and Service Management · doi:10.1109/tnsm.2018.2808352 · Cited by 23
“While much of the recent work has focused on hardware acceleration using GPUs [20], [44], FPGAs [45], [46], and ASICs [26], [47], parallelism has being shown to be critical”
BackgroundMethodInfluentialA High-efficiency FPGA-based Accelerator for Convolutional Neural Networks using Winograd Algorithm
Huang, Y., Shen, J., Wang, Z., Wen, M., & Zhang, C. (2018) · Journal of Physics Conference Series · doi:10.1088/1742-6596/1026/1/012019 · Cited by 23
“Synthesizing the overall performance and resource utilization, the design for (4 4,3 3) F achieves better performance.”
Self-citationA Multi-Level-Optimization Framework for FPGA-Based Cellular Neural Network Implementation
Liu, Z., Luo, S., Xu, X., Shi, Y., & Zhuo, C. (2018) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3273957 · Cited by 23
“As in Reference [33], computational density is defined as the average GOPS per area unit (GOPS/LUTs).”
BackgroundFune: An FPGA Tuning Framework for CNN Acceleration
Xiao, Q., & Liang, Y. E. (2019) · IEEE Design and Test · doi:10.1109/mdat.2019.2908549 · Cited by 23
“Similarly, 1 × 1 × 7 operation parallelism and 64 × 1 × 1 output parallelism are explored by the CU from [10], leading to 57.”
BackgroundWinograd-Based Real-Time Super-Resolution System on FPGA
Shi, B., Tang, Z., Luo, G., & Jiang, M. F. (2019) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/icfpt47387.2019.00083 · Cited by 23
“Among the works of FPGA accelerator designs[8], [9], they mainly explore the computing parallelism, pipelines and the data organization.”
BackgroundDynamic Sharing in Multi-accelerators of Neural Networks on an FPGA Edge Device
Ting, H.-Y., Giyahchi, T., Sani, A. A., & Bozorgzadeh, E. (2020) · IEEE International Conference on Application-specific Systems, Architectures and Processors (ASAP) · doi:10.1109/asap49362.2020.00040 · Cited by 23
Artificial Intelligence Accelerators Based on Graphene Optoelectronic Devices
Gao, W., Yu, C., & Chen, R. (2021) · Advanced Photonics Research · doi:10.1002/adpr.202100048 · Cited by 23
Chen, X., Chen, Q., Tian, M., Ji, M., Zou, C., Wang, X., & Wang, B. (2017) · IEEE International Conference on Parallel and Distributed Systems (ICPADS) · doi:10.1109/icpads.2017.00034 · Cited by 23
“Accordingly, relocating the data reuse to the lower storage level as much as possible delicately is one of the most common and effective ways to reduce energy consumption [9–12].”
MethodEnergy‐based tuning of convolutional neural networks on multi‐GPUs
Castro, F. M., Guil, N., Marín-Jiménez, M. J., Pérez-Serrano, J., & Ujaldón, M. (2018) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.4786 · Cited by 23
“In addition, we show the Computation to Communication ratio [48] defined as: CTC = N op N d where N op is the total number of operations and N d is the amount of data read.”
Hardware Compilation of Deep Neural Networks: An Overview
Zhao, R., Liu, S., Ng, H.-C., Wang, E., Davis, J. J., Niu, X., Wang, X., Shi, H.-F., Constantinides, G. A., Cheung, P. Y. K., et al. (2018) · IEEE International Conference on Application-specific Systems, Architectures and Processors (ASAP) · doi:10.1109/asap.2018.8445088 · Cited by 23
“In some literature, the authors aimed to design templates with high scalability and flexibility [12], [36]–[40], while others applied them within design space explorations to maximise performance [41]– [43].”
BackgroundMethodInfluentialTraining Binary Weight Networks via Semi-Binary Decomposition
Hu, Q., Li, G., Wang, P., Zhang, Y., & Cheng, J. (2018) · Lecture notes in computer science · doi:10.1007/978-3-030-01261-8_39 · Cited by 23
“For fair comparison, we adopt the same platform and working frequency, and restrict the usage of on-chip computing resources (LUTs and FFs) as the same level as in [29].”
BackgroundMethodInfluentialKelp: QoS for Accelerated Machine Learning Systems
Zhu, H., Lo, D., Cheng, L., Govindaraju, R., Ranganathan, P., & Erez, M. (2019) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca.2019.00036 · Cited by 22
“Many other ASIC neural network accelerator designs (e.g., [15], [16], [17], [18]) and optimizations (e.g., [19], [20], [21], [22]) have been proposed while other work explores the option of using FPGA-based solutions (e.g., [55], [56], [57], [58]).”
BackgroundEfficient Algorithms for Embedded Tactile Data Processing
Younes, H., Alameh, M., Ibrahim, A., Rizk, M., & Valle, M. (2022) · River Publishers eBooks · doi:10.1201/9781003338062-6 · Cited by 22
Liu, Q., Setter, J., Huff, D., Strange, M., Feng, K., Horowitz, M., Raina, P., & Kjølstad, F. (2022) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3572908 · Cited by 22
“To efficiently execute DNNs, Zhang et al. [ 53 ] optimize DNN data blocking using double buffer structures and synthesize a pipelined FPGA accelerator from Caffe [ 18 ].”
MethodALT: Breaking the Wall between Data Layout and Loop Optimizations for Deep Learning Compilation
Xu, Z., Xu, J., Peng, H., Wang, W., Wang, X., Wan, H., Dai, H., Xu, Y., Cheng, H., Wang, K., et al. (2023) · European Conference on Computer Systems · doi:10.1145/3552326.3587440 · Cited by 23
“Many systems try to improve the performance with layout transformation [10, 14, 21, 40, 42, 43, 52, 78, 82].”
BackgroundMethodA High-Performance Accelerator for Real-Time Super-Resolution on Edge FPGAs
Liu, H., Qian, Y., Liang, Y., Zhang, B., Liu, Z., He, T., Zhao, W., Lu, J., & Yu, B. (2024) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3652855 · Cited by 22
“The roofline model [37, 38] suggests that the actual performance of a system can be constrained by either the computation roof or the communication to computation ratio.”
BackgroundLiu, Y., Ma, Y., Zhang, B., Liu, L., Wang, J., & Tang, S. (2024) · Microelectronics Journal · doi:10.1016/j.mejo.2024.106197 · Cited by 22
Building Your Own Trusted Execution Environments Using FPGA
Armanuzzaman, M., Sadeghi, A., & Zhao, Z. (2024) · ACM Asia Conference on Computer and Communications Security (ASIA CCS) · doi:10.1145/3634737.3637644 · Cited by 22
“Various FPGA-based application-speci � c accelerators, such as deep neural networks [27–29], classic and post-quantum cryptographic algorithms [30, 31], Memcached [32], have been proposed and deployed.”
MethodDANoC: An Efficient Algorithm and Hardware Codesign of Deep Neural Networks on Chip
Zhou, X., Li, S., Tang, F., Hu, S., Lin, Z. W., & Zhang, L. (2017) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2017.2717442 · Cited by 22
“[28] quantitatively analyzed”
Piyasena, D., Wickramasinghe, R., Paul, D., Lam, S.-K., & Wu, M. (2019) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2019.00063 · Cited by 22
“FPGA based CNN accelerators are commonly implemented in the form of a time-shared uniform accelerator which processes layers sequentially [19], [20].”
MethodSurvey on memory management techniques in heterogeneous computing systems
Hazarika, A., Poddar, S., & Rahaman, H. (2019) · IET Computers & Digital Techniques · doi:10.1049/iet-cdt.2019.0092 · Cited by 22
“Zhang et al. [85] proposed a novel architecture for the CNN accelerator on the FPGA board to overcome this problem by using the roofline model [86].”
MethodBinary convolutional neural network acceleration framework for rapid system prototyping
Xu, Z., & Cheung, R. C. C. (2020) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2020.101762 · Cited by 22
Henry, G., Palangpour, P., Thomson, M., Gardner, J., Arden, B., Donahue, J., Houck, K., Johnson, J., O’Brien, K. H., Petersen, S., et al. (2020) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1109/isca45697.2020.00013 · Cited by 22
“Due to the flexible nature of FPGAs, a large body of research has focused on developing FPGA-based DL accelerators [25] [21] [24] [17] [28] [19].”
BackgroundDinelli, G., Meoni, G., Rapuano, E., Pacini, T., & Fanucci, L. (2020) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2020.3015294 · Cited by 22
“Many studies such as [13] show as FPGAs overcome CPUs both for what concerns power consumption and timing performance.”
BackgroundFPGA based neural network accelerators
Kim, J.-Y. (2020) · Advances in computers · doi:10.1016/bs.adcom.2020.11.002 · Cited by 22
Sateesan, A., Sinha, S., Smitha, K. G., & Vinod, A. P. (2021) · Neural Processing Letters · doi:10.1007/s11063-021-10458-1 · Cited by 23
Enhancing the Scalability of Multi-FPGA Stencil Computations via Highly Optimized HDL Components
Reggiani, E., Sozzo, E. D., Conficconi, D., Natale, G., Moroni, C., & Santambrogio, M. D. (2021) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3461478 · Cited by 22
SAMO: Optimised Mapping of Convolutional Neural Networks to Streaming Architectures
Montgomerie-Corcoran, A., Yu, Z., & Bouganis, C.-S. (2022) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl57034.2022.00069 · Cited by 22
“Unlike Systolic Array architectures where the design space exploration can be carried out by brute force enumeration with the help of an analytical resource model [8], or a roofline model [9], the Streaming Architecture requires a more informed method for efficiently exploring the design space.”
MethodNekooei, A., & Safari, S. (2022) · Neural Networks · doi:10.1016/j.neunet.2022.02.024 · Cited by 22
An OpenCL-Based FPGA Accelerator for Faster R-CNN
An, J., Zhang, D., Xu, K., & Wang, D. (2022) · Entropy · doi:10.3390/e24101346 · Cited by 23
“Furthermore, FPGA-based CNN hardware accelerator designs [10–15] are rapidly developing due to their reconfigurability and fast development time, especially when FPGA vendors provide high-level synthesis (HLS) tools.”
Lossy Compression for Embedded Computer Vision Systems
Guo, L., Zhou, D., Zhou, J., Kimura, S., & Goto, S. (2018) · IEEE Access · doi:10.1109/access.2018.2852809 · Cited by 22
AutoDNNchip: An Automated DNN Chip Predictor and Builder for Both FPGAs and ASICsPreprint
Xu, P., Zhang, X., Hao, C., Zhao, Y. K., Zhang, Y., Wang, Y., Li, C., Guan, Z., Chen, D., & Lin, Y. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2001.03535 · Cited by 21
“Recently, we have seen intensive studies on DNN accelerators in hardware, which attempt to take advantage of different hardware design styles, such as GPUs, FPGAs, and ASICs, to improve the speed and efficiency of DNN inference and training [10–22].”
BackgroundMethodColbert, I., Daly, J., Kreutz-Delgado, K., & Das, S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.00294 · Cited by 22
“Any solution to the left of the peak sustainable bandwidth slope requires a higher bandwidth than the FPGA can sustain [17].”
BackgroundMethodInfluentialLearning on Hardware: A Tutorial on Neural Network Accelerators and Co-ProcessorsPreprint
Baischer, L., Wess, M., & TaheriNejad, N. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.09252 · Cited by 22
“However, the high precision results in a low throughput of 61.62GOPS [58].”
BackgroundSnowflake: A Model Agnostic Accelerator for Deep Convolutional Neural NetworksPreprint
Gokhale, V., Zaidy, A., Chang, A. X. M., & Culurciello, E. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1708.02579 · Cited by 22
“[27] use a 32-bit floating point accelerator.”
Ning, S., Zhu, H., Feng, C., Gu, J., Pan, D. Z., & Chen, R. T. (2025) · Optica · doi:10.1364/optica.559604 · Cited by 21
“To efficiently execute MVMs, hardware AI accelerators—including, but not limited to, graphical processing units (GPUs), field-programmable gate arrays (FPGAs), [13][14] and application-specific integrated circuits (ASICs) [15] —have been widely developed by both industry and academia.”
Zheng, J., Lv, Z., Li, D., Lu, C., Shen, Z., Chen, H., Huang, X., Huang, J., Chen, D., & Zhang, J. (2025) · Measurement · doi:10.1016/j.measurement.2025.118122 · Cited by 21
An Efficient CNN Architecture for Image Classification on FPGA Accelerator
Mujawar, S., Kiran, D., & Ramasangu, H. (2018) · International Conference on Advances in Electronics Computers and Communications · doi:10.1109/icaecc.2018.8479517 · Cited by 21
“[10] proposed tile computations for CNN and implemented CNN on Virtex-7 FPGA.”
MethodSimple and Robust Deep Learning Approach for Fast Fluorescence Lifetime Imaging
Wang, Q., Li, Y., Xiao, D., Zang, Z., Jiao, Z., Chen, Y., & Li, D. (2022) · Sensors · doi:10.3390/s22197293 · Cited by 21
“CMM [20,23] offers fast analysis, but it can only provide intensity-weighted average lifetimes [14].”
BackgroundMethodLal, S., Chanchal, A. K., Kini, J. R., & Upadhyay, G. K. (2024) · Multimedia Tools and Applications · doi:10.1007/s11042-023-17895-1 · Cited by 21
“A kernel design methodology based on OpenCL is proposed in [21] to overcome bandwidth requirement but the performance of the first convolutional layer fails to be credible.”
Unleashing Network/Accelerator Co-Exploration Potential on FPGAs: A Deeper Joint Search
Lou, W., Gong, L., Wang, C., Qian, J., Wang, X., Li, C., & Zhou, X. (2024) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2024.3391688 · Cited by 21
“…accelerators due to their reconfigurability and customizability for quickly providing optimized support for different NNs. Previous studies [25], [26] have explored micro-architecture design, such as parallelization and tiling, where a single computing core with fixed dataflow is reused for…”
BackgroundBordawekar, R. R., Blainey, B., & Puri, R. (2014) · Synthesis lectures on computer architecture · doi:10.1007/978-3-031-01749-0 · Cited by 21
Bit-Width Based Resource Partitioning for CNN Acceleration on FPGA
Guo, J., Yin, S., Ouyang, P., Liu, L., & Wei, S. (2017) · IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM) · doi:10.1109/fccm.2017.13 · Cited by 21
“Experiment (1-3, 2-3, 3-3) show the results of our FPGA bit-width partitioning design.”
BackgroundCaffePresso: Accelerating Convolutional Networks on Embedded SoCs
Hegde, G., Siddhartha, & Kapre, N. (2017) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3105925 · Cited by 21
“Unlike solutions (FPGAs [Ovtcharov et al. 2015; Zhang et al. 2015; Gokhale et al. 2014], GPUs [Wu et al.”
BackgroundMethodInfluentialSqueezeJet: High-Level Synthesis Accelerator Design for Deep Convolutional Neural Networks
Mousouliotis, P., & Petrou, L. P. (2018) · Lecture notes in computer science · doi:10.1007/978-3-319-78890-6_5 · Cited by 21
“…those which accelerate only the conv layer and those which accelerate two or more layer types of a DCNN. Conv layer accelerators: Zhang et al. [4] designed an architecture template for the conv layer using loop tiling, loop arrangement based on data dependencies, computation optimizations…”
BackgroundMethodInfluentialZhang, J. J., Raj, P., Zarar, S., Ambardekar, A., & Garg, S. (2019) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3358178 · Cited by 21
“This includes work on optimizing the SA architecture including the shape of the SA core and buffer sizes [16, 42], optimized memory schedules [10], and techniques to reduce power consumption of the SA core [40].”
BackgroundDistributed Inference over Decision Tree Ensembles on Clusters of FPGAs
Owaida, M., Kulkarni, A., & Alonso, G. (2019) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3340263 · Cited by 21
TensorFlow to Cloud FPGAs: Tradeoffs for Accelerating Deep Neural Networks
Hadjis, S., & Olukotun, K. (2019) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2019.00064 · Cited by 21
A Throughput-Latency Co-Optimised Cascade of Convolutional Neural Network Classifiers
Kouris, A., Venieris, S. I., & Bouganis, C.-S. (2020) · Design, Automation, and Test in Europe · doi:10.23919/date48585.2020.9116248 · Cited by 21
“[28]), we employ a resource sharing approach targeting to fit an {LPU,HPU} pair concurrently on the target device.”
MethodFiC-RNN: A Multi-FPGA Acceleration Framework for Deep Recurrent Neural Networks
Sun, Y., & Amano, H. (2020) · IEICE Transactions on Information and Systems · doi:10.1587/transinf.2020pap0003 · Cited by 21
NullaNet Tiny: Ultra-low-latency DNN Inference Through Fixed-function Combinational Logic
Nazemi, M., Fayyazi, A., Esmaili, A., Khare, A., Shahsavani, S. N., & Pedram, M. (2021) · IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM) · doi:10.1109/fccm51124.2021.00053 · Cited by 21
Enabling energy-efficient DNN training on hybrid GPU-FPGA accelerators
He, X. J., Liu, J., Xie, Z., Chen, H., Chen, G., Zhang, W., & Li, D. M. (2021) · International Conference on Supercomputing · doi:10.1145/3447818.3460371 · Cited by 21
“Much success has been achieved in using FPGAs to accelerate machine learning inference [35, 45, 48, 70, 72], although limited works have explored the use of FPGAs for training [21, 31].”
BackgroundMore Is Less: Model Augmentation for Intermittent Deep Inference
Kang, C.-K., Mendis, H. R., Lin, C., Chen⋆, M.-S., & Hsiu, P.-C. (2022) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3506732 · Cited by 21
“Early FPGA-based custom accelerators focused on exploiting data reuse [11, 13, 21, 59, 73, 74], resulting in accelerator designs optimized for a particular computation order.”
BackgroundFPGA-Based Hardware Accelerator on Portable Equipment for EEG Signal Patterns Recognition
Xie, Y., Majoros, T., & Oniga, S. (2022) · Electronics · doi:10.3390/electronics11152410 · Cited by 21
Increasing FPGA Accelerators Memory Bandwidth With a Burst-Friendly Memory Layout
Ferry, C., Yuki, T., Derrien, S., & Rajopadhye, S. V. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3201494 · Cited by 21
“The impact of tiling on performance is such that finding the best tile size and shape for specific classes of programs, such as matrix multiplication [11] or convolution [9], is a trending topic in compiler research.”
BackgroundPNeuro: A scalable energy-efficient programmable hardware accelerator for neural networks
Carbon, A., Philippe, J., Bichler, O., Schmit, R., Tain, B., Briand, D., Ventroux, N., Paindavoine, M., & Brousse, O. (2018) · Design, Automation, and Test in Europe · doi:10.23919/date.2018.8342165 · Cited by 21
“Recent works [5], [6] demonstrate their capacity to adapt their structure to dataflow and massively parallel implementations of DNNs.”
BackgroundAn Asynchronous Energy-Efficient CNN Accelerator with Reconfigurable Architecture
Chen, W., Wu, H., Wei, S., He, A., & Chen, H. (2018) · IEEE Asian Solid-State Circuits Conference (A-SSCC) · doi:10.1109/asscc.2018.8579304 · Cited by 21
Compact Convolutional Neural Network Accelerator for IoT Endpoint SoC
Ge, F., Wu, N., Xiao, H., Zhang, Y., & Zhou, F. (2019) · Electronics · doi:10.3390/electronics8050497 · Cited by 21
“As stated in [5,6], most of the network structures in CNN are deployed and implemented by hardware.”
BackgroundPareto Optimal Design Space Exploration for Accelerated CNN on FPGA
Reggiani, E., Rabozzi, M., Nestorov, A. M., Scolari, A., Stornaiuolo, L., & Santambrogio, M. D. (2019) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw.2019.00028 · Cited by 21
“While [4] and [9] employ manual analytical models to perform the CNN DSE, the authors in [5], [10]–[12] propose automatic tool flows to infer the best network configuration.”
Method
Metadata
- Authors
- Zhang, C., Li, P., Sun, G., Guan, Y., Xiao, B., & Cong, J.
- Affiliations
- Peking University (CN)
- University of California, Los Angeles (US)
- Venue
- Proceedings of the 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays
- Publisher
- ACM
- Published
- 2015-02-22
- Pages
- 161-170
- OpenAlex
- W2094756095
- Cited by
- 2,158
- References
- 17
- Citations by year
- 2026: 78 · 2025: 116 · 2024: 145 · 2023: 162 · 2022: 233 · 2021: 253 · 2020: 258 · 2019: 303 · 2018: 281 · 2017: 221 · 2016: 90 · 2015: 15
- FWCI
- 133.92
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) (publishedVersion)
- Funders
- C-FAR
- NSF China (61202072)
- National High Technology Research and Development Program of China (2012AA010902)
- RFDP (20110001110099)
- Topics
- Physical Sciences › Computer Science › Hardware and Architecture › Embedded Systems Design Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Embedded Systems and FPGA Applications
- Keywords
- FPGA accelerator, deep convolutional neural networks, roofline model, memory bandwidth, loop tiling, loop transformation, FPGA resource utilization, image recognition
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.