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 901–1,000 of 1,002
VHDL generator for a high performance convolutional neural network FPGA-based accelerator
Hamdan, M. K. A., & Rover, D. T. (2017) · International Conference on ReConFigurable Computing and FPGAs (ReConFig) · doi:10.1109/reconfig.2017.8279827 · Cited by 17
“Indeed, many CNN accelerators have been proposed for different purposes and with different techniques and methodologies [5][6] [7][8][9].”
BackgroundEnergy-Performance Considerations for Data Offloading to FPGA-Based Accelerators Over PCIe
Mbakoyiannis, D., Tomoutzoglou, O., & Kornaros, G. (2018) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3180263 · Cited by 17
“Hence, similar to the roofline model, commonly used to depict the peak platform performance and potential bottlenecks [22, 33], we plot the gathered metrics for the various workloads in the roofline graph on the right.”
MethodOn-FPGA Training with Ultra Memory Reduction: A Low-Precision Tensor MethodPreprint
Zhang, K., Hawkins, C., Zhang, X., Hao, C., & Zhang, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2104.03420 · Cited by 17
“…limited hardware resources have been well studied at both the algo-rithm (LeCun et al., 1990; Neklyudov et al., 2017; Lebedev et al., 2014; Hinton et al., 2015; Han et al., 2015; Zhou et al., 2017) and hardware levels (Chen et al., 2016; 2019; Zhang et al., 2015; Hao et al., 2019; Xu et al., 2020).”
BackgroundAn Efficient FPGA-Based Accelerator for Swin TransformerPreprint
Liu, Z., Yin, P., & Ren, Z. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2308.13922 · Cited by 18
“Researchers in [6], [7], [8] have extensively optimized loop methodologies and dataflow in accelerators to maximize FPGA resource utilization and to get the best performance.”
BackgroundXu, S., Wong, K., Du, Y., Hong, H., Chae, C., Liu, B., & Tong, K. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3637390 · Cited by 17
“According to the roofline model [35], the primary factors limiting inference latency in FPGA-based accelerators are computational throughput and the bandwidth constraints between on-chip and off-chip memory.”
Reconfigurable Virtual Memory for FPGA-Driven I/O
Landgraf, J., Giordano, M., Yoon, E., & Rossbach, C. J. (2023) · International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) · doi:10.1145/3582016.3582048 · Cited by 16
“FPGAs’ tight coupling of memory and logic makes them competitive for a wide variety of workloads, including communication [23, 51, 55, 92, 108], databases [24, 57, 74], image processing [81], finance [52, 69], graph processing [33, 82], and machine learning [29, 97, 104, 121].”
Real-time detection of track fasteners based on object detection and FPGA
Xiao, T. Y., Xu, T., & Wang, G. (2023) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2023.104863 · Cited by 16
Edavoor, P. J., Samantaray, A. K., & Rahulkar, A. D. (2024) · e-Prime - Advances in Electrical Engineering Electronics and Energy · doi:10.1016/j.prime.2024.100546 · Cited by 16
Li, Q., Wang, Y., Meng, L., Qin, Y., & Tang, B. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3548775 · Cited by 16
“Current model lightweight techniques focus on three main aspects: lightweight modeling [20], [21], model compression [22], [23], and acceleration techniques [24], [25].”
LLM-NPU: Towards Efficient Foundation Model Inference on Low-Power Neural Processing Units
Raha, A., Kundu, S., Sridhar, S. N., Kundu, S., Ghosh, S. K., Palla, A., Das, A., Crews, D., & Mathaikutty, D. A. (2025) · IEEE International Conference on Omni-layer Intelligent Systems (COINS) · doi:10.1109/coins65080.2025.11125797 · Cited by 16
“Apart from model optimizations, various compiler-level optimizations such as operator fusion and tiling operations [17] have translated to efficient data reuse and reduced memory operations.”
Using Data Compression for Optimizing FPGA-Based Convolutional Neural Network Accelerators
Guan, Y., Xu, N., Zhang, C., Yuan, Z., & Cong, J. (2017) · Lecture notes in computer science · doi:10.1007/978-3-319-67952-5_2 · Cited by 16
“Using the method provided in [16], we can find a point (X in Fig.”
BackgroundMethodSelf-citationMemory Optimization Techniques for FPGA based CNN Implementations
Shahshahani, M., Goswami, P., & Bhatia, D. K. (2018) · IEEE Dallas Circuits and Systems Conference (DCAS) · doi:10.1109/dcas.2018.8620112 · Cited by 16
“Design space exploration of supported platform in [11].”
BackgroundMethodInfluentialShu, C., Pang, W., Liu, H., & Lu, S. (2019) · IEEE International Conference on Signal and Image Processing (ICSIP) · doi:10.1109/siprocess.2019.8868502 · Cited by 16
“[11] [12] [13] Ours”
A streaming architecture for Convolutional Neural Networks based on layer operations chaining
Arredondo-Velázquez, M., Díaz-Carmona, J., Torres-Huitzil, C., Padilla-Medina, J. A., & Prado-Olivarez, J. (2020) · Journal of Real-Time Image Processing · doi:10.1007/s11554-019-00938-y · Cited by 16
“In the first strategy, authors are focused on achieving a processing acceleration in all network layers using high parallelism level in the convolver, whereas the convolutional layers are sequentially computed [25–28].”
A Real-time Online Aircraft Neural Network System
Zhang, Y., Zhao, Q., Tao, L., Cao, J., Wei, M., & Zhang, X. (2019) · doi:10.1109/iwofc48002.2019.9078469 · Cited by 16
“Paper [2][17] Paper[3][9] This article”
BackgroundBinary Precision Neural Network Manycore Accelerator
Hosseini, M., & Mohsenin, T. (2021) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3423136 · Cited by 16
“5 BNN Implementation Flow As mentioned in the previous Section for the inference of the BNN, we follow the output channel tiling scheme, in which the computation to communication (CTC) ratio as compared to other tiling schemes is minimum [43], in an intra-cluster level.”
MethodAutomated CNN back-propagation pipeline generation for FPGA online training
Mazouz, A., & Bridges, C. (2021) · Journal of Real-Time Image Processing · doi:10.1007/s11554-021-01147-2 · Cited by 16
“…the hardware implementation of forward pass CNNs as inference engines and accelerators, there is plenty of research done to map the CNN forward pass unto FPGAs for embedded inference [2–18], in contrast, there is a clear lack of work in the areas of online deployment and training on FPGAs.”
BackgroundParallel Implementation of CNN on Multi-FPGA Cluster
Fukushima, Y., Iizuka, K., & Amano, H. (2021) · IEEE International Symposium on Embedded Multicore/Many-core Systems-on-Chip (MCSoC) · doi:10.1109/mcsoc51149.2021.00019 · Cited by 16
The Impact of Data Representationson Hardware Based MLP Network Implementation
(2014) · Acta Polytechnica Hungarica · doi:10.12700/aph.15.1.2018.2.4 · Cited by 16
Li, T., He, B., & Zheng, Y. (2023) · Applied Sciences · doi:10.3390/app13021003 · Cited by 16
DT-CGRA: Dual-track coarse-grained reconfigurable architecture for stream applications
Fan, X., Li, H., Cao, W., & Wang, L. (2016) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2016.7577309 · Cited by 16
“However, the proposed DT-CGRA outperforms [19] with 7.”
MethodResultInfluentialTile size selection for optimized memory reuse in high-level synthesis
Liu, J., Wickerson, J., & Constantinides, G. A. (2017) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.23919/fpl.2017.8056810 · Cited by 16
“Especially for CNN accelerators, exhaustive enumeration (as used in [11]) and random enumeration (as used in [12]) are used for design space exploration.”
MethodBit-width Adaptive Accelerator Design for Convolution Neural Network
Guo, J., Yin, S., Ouyang, P., Tu, F., Tang, S., Liu, L., & Wei, S. (2018) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2018.8351666 · Cited by 16
“For this reason, various accelerators based on FPGA, GPU, and even ASIC have been proposed to improve performance of CNNs [15] [8] [5].”
InfluentialLiu, X., Kim, D. H., Wu, C., & Chen, D. (2018) · doi:10.1145/3225209.3225214 · Cited by 16
“In [14], an FPGA based CNN network accelerator is proposed.”
BackgroundMethodInfluentialFPGA-based CNN Processor with Filter-Wise-Optimized Bit Precision
Maki, A., Miyashita, D., Nakata, K., Tachibana, F., Suzuki, T., & Deguchi, J. (2018) · IEEE Asian Solid-State Circuits Conference (A-SSCC) · doi:10.1109/asscc.2018.8579342 · Cited by 16
QRS Detection in ECG Signal with Convolutional Network
Silva, P., Luz, E. J. D. S., Wanner, E. F., Menotti, D., & Moreira, G. J. P. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-13469-3_93 · Cited by 16
Accelerating Stochastic Gradient Descent Based Matrix Factorization on FPGA
Zhou, S., Kannan, R., & Prasanna, V. K. (2020) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2020.2974744 · Cited by 16
“FPGA has been widely used to accelerate the inference process of machine learning applications [3], [4], [5], [6], [7], but using FPGA to accelerate the training process remains a challenging research area [8], [9].”
BackgroundMethodReview of prominent strategies for mapping CNNs onto embedded systems
Arredondo-Velázquez, M., Díaz-Carmona, J., Gutiérrez, A. I. B., & Torres-Huitzil, C. (2020) · IEEE Latin America Transactions · doi:10.1109/tla.2020.9082927 · Cited by 16
“As´ı, la convoluci´on com´unmente se realiza mediante un ´arbol de multiplicadores de la misma dimensi´on que el kernel de convoluci´on, algunos ejemplos de este modo de operaci´on se reportan en [94], [106].”
BackgroundInfluentialKim, S.-H., Lee, C., & Youn, C. (2020) · IEEE Access · doi:10.1109/access.2020.3033771 · Cited by 16
unzipFPGA: Enhancing FPGA-based CNN Engines with On-the-Fly Weights Generation
Venieris, S. I., Fernández-Marqués, J., & Lane, N. D. (2021) · IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM) · doi:10.1109/fccm51124.2021.00027 · Cited by 16
“Both a) and b) are parametrised with tile sizes 〈TR, TP , TC〉 and roofline modelling [19] is used to obtain the highest throughput configuration for the target CNN-FPGA pair.”
BackgroundMethodGeneral Reuse-Centric CNN Accelerator
Cicek, N. M., Ning, L., Öztürk, Ö., & Shen, X. (2021) · IEEE Transactions on Computers · doi:10.1109/tc.2021.3064608 · Cited by 16
“Among them, FPGA [1], [4], [33], [34], [35], [36], [37] is well studied and shows great potential on accelerating DNN inference.”
BackgroundKwon, J., Kong, J., & Munir, A. (2021) · IET Computers & Digital Techniques · doi:10.1049/cdt2.12038 · Cited by 16
“[5–9] that have the costs similar to the proposed accelerator may distort the results due to suboptimal or different implementations compared to those in Refs.”
BackgroundMethodResultInfluentialAn Efficient Implementation of Convolutional Neural Network With CLIP-Q Quantization on FPGA
Wei, C., Lin, I.-C., & Shih, Y.-Y. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3193031 · Cited by 16
“High-level synthesis was used to generate the design with the help of roofline model in [33].”
MethodA Programmable Hardware Accelerator for Simulating Dynamical Systems
Kung, J., Long, Y., Kim, D.-H., & Mukhopadhyay, S. (2017) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3079856.3080252 · Cited by 16
“The dataflow scheme can be grouped into four categories: no local reuse (NLR [45]), weight stationary (WS [4]), output stationary (OS [11, 14, 34]), and row stationary (RS [6]).”
BackgroundMethodPandey, S., Siddhu, L., & Panda, P. R. (2023) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3630012 · Cited by 16
“Such a treatment could incur unnecessary performance loss as diferent applications have diferent memory-access behavior and contribute diferently to memory temperature.”
BackgroundLow-Bit Mixed-Precision Quantization and Acceleration of CNN for FPGA Deployment
Wang, J., He, Z., Zhao, H., & Liu, R. (2024) · IEEE Transactions on Emerging Topics in Computational Intelligence · doi:10.1109/tetci.2024.3510295 · Cited by 15
“Based on work [29], Ma et al. [30] further explored parallelism on width and height dimensions of input features, and expand the designspacefor morefeasiblesolutions.”
Zheng, J., Lv, Z., Li, D., Lu, C., Zhang, Y., Fu, L. P., Huang, X., Huang, J., Chen, D., & Zhang, J. (2025) · Electronics · doi:10.3390/electronics14091704 · Cited by 15
“In most CNN architectures, ping-pong buffering is utilized to store data in two separate buffers: an input buffer and an output buffer [26].”
Dynamic Approximation with Feedback Control for Energy-Efficient Recurrent Neural Network Hardware
Kung, J., Kim, D.-H., & Mukhopadhyay, S. (2016) · International Symposium on Low Power Electronics and Design (ISLPED) · doi:10.1145/2934583.2934626 · Cited by 15
“Hardware accelerators are being investigated to improve power-efficiency of NN computations [3].”
BackgroundArtificial intelligence enhances the performance of chaotic baseband wireless communication
Ren, H., Yin, H., Zhao, H., Bai, C., & Grebogi, C. (2021) · IET Communications · doi:10.1049/cmu2.12162 · Cited by 15
“For the CNN computation in FPGA, there is some reported example in [36], where VC707 FPGA chip is used to achieve the peak performance of 61.”
MethodDesigning Deep Learning Hardware Accelerator and Efficiency Evaluation
Qi, Z., Chen, W., Naqvi, R. A., & Siddique, K. (2022) · Computational Intelligence and Neuroscience · doi:10.1155/2022/1291103 · Cited by 15
“At present, most of the research focuses on the repetition part of matrix operation in FPGA computing optimization [5, 6], but few on the improvement of activation function optimization, so this is a field to be excavated.”
BackgroundHigh-Performance Big Data Computing
Panda, D. K. D., Lü, X., & Shankar, D. (2022) · The MIT Press eBooks · doi:10.7551/mitpress/11451.001.0001 · Cited by 15
John, C., Sahoo, J., Madhavan, M., & Mathew, O. K. (2023) · Current Bioinformatics · doi:10.2174/1574893618666230320103421 · Cited by 15
MOCHA: Morphable Locality and Compression Aware Architecture for Convolutional Neural Networks
Jafri, S. M. A. H., Hemani, A., Paul, K., & Abbas, N. (2017) · IEEE International Parallel and Distributed Processing Symposium (IPDPS) · doi:10.1109/ipdps.2017.59 · Cited by 15
“In FPGA domain, the CNNs are mapped using a combination of look-up-tables and the DSPs [2], [3].”
BackgroundMethodInfluentialTNPU: an efficient accelerator architecture for training convolutional neural networks
Li, J., Yan, G., Lü, W., Jiang, S., Gong, S., Wu, J., Yan, J., & Li, X. (2019) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3287624.3287641 · Cited by 15
“Work [16] focuses on the balance between computing resource and memory bandwidth for CVLs.”
BackgroundMinimizing Classification Energy of Binarized Neural Network Inference for Wearable Devices
Hosseini, M., Paneliya, H., Kallakuri, U., Khatwani, M., & Mohsenin, T. (2019) · International Symposium on Quality Electronic Design (ISQED) · doi:10.1109/isqed.2019.8697574 · Cited by 15
“This scheme is shown to result in the best computation to memory communication ratio [24].”
MethodA Solution to Optimize Multi-Operand Adders in CNN Architecture on FPGA
Farrukh, F. U. D., Xie, T., Zhang, C., & Wang, Z. (2019) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2019.8702777 · Cited by 15
“Due to these benefits, it draws attention of many researchers to put their efforts to optimize CNN implementation on FPGA [7-9].”
BackgroundMethodSelf-citationLi, S., Sun, K., Luo, Y., Yadav, N., & Choi, K. (2020) · Electronics · doi:10.3390/electronics9050832 · Cited by 15
“There were some works [17,20] that applied the roofline model to design the trade-off between computing throughput and required memory bandwidth to maximize the utilization of FPGA resources.”
MethodInfluentialAutomated Design Space Exploration for Optimized Deployment of DNN on Arm Cortex-A CPUs
de Prado, M., Mundy, A., Saeed, R., Denna, M., Pazos, N., & Benini, L. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3046568 · Cited by 15
Song, Z., Wang, J., Li, T., Jiang, L., Ke, J., Liang, X., & Jing, N. (2020) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac18072.2020.9218566 · Cited by 15
“Some accelerators are proposed to maximize the storage capacity, such as a FPGAbased accelerator [19] and Diannao [20].”
BackgroundHardware accelerator systems for embedded systems
Song, W. J. (2020) · Advances in computers · doi:10.1016/bs.adcom.2020.11.004 · Cited by 16
Nider, J., & Fedorova, A. (2021) · Workshop on Hot Topics in Operating Systems · doi:10.1145/3458336.3465291 · Cited by 15
“Accelerators for many applications such as image recognition [45], computer vision [22], key-value stores [14, 30], data warehouses [26], big data [10, 18, 44], deep learning [12], neural networks [23] (and many more) are commonly used to reduce overall system cost and increase performance orders of magnitude beyond t…”
BackgroundMoRS: An Approximate Fault Modeling Framework for Reduced-Voltage SRAMs
Yüksel, İ. E., Salami, B., Ergin, O., Ünsal, O., & Kestelman, A. C. (2021) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2021.3120073 · Cited by 15
Spiking neural networks — Algorithms, hardware implementations and applications
Kulkarni, S., Babu, A. V., & Rajendran, B. (2017) · IEEE International Midwest Symposium on Circuits and Systems (MWSCAS) · doi:10.1109/mwscas.2017.8052951 · Cited by 15
“An FPGA-based implementation of an SNN with two million neurons was shown to achieve a 38× speedup compared to a serial implementation on a CPU [40].”
InfluentialFPGA-based accelerator for losslessly quantized convolutional neural networks
Sit, M.-K., Kazami, R., & Amano, H. (2017) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/fpt.2017.8280164 · Cited by 15
“Compared to GPUs, FPGAs have much fewer floating point units and less on-chip memory buffers, which limits their capability to accelerate the floating point precision models.”
MethodAn efficient reconfigurable hardware accelerator for convolutional neural networks
Ansari, A., Gunnam, K., & Ogunfunmi, T. (2017) · Asilomar Conference on Signals, Systems and Computers (ACSSC) · doi:10.1109/acssc.2017.8335571 · Cited by 15
“The work done in [6], explores the communication to computation ratio, and the computation roofline space for each layer to determine the computational performance with optimal unroll factors < Tm, Tn >, where Tm and Tn are the tile size of the outputs and inputs of the computation engine respectively as seen in Table…”
BackgroundMethodInfluentialReducing the Computation Load of Convolutional Neural Networks through Gate Classification
Disabato, S., & Roveri, M. (2018) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2018.8489276 · Cited by 15
“The problem of reducing the complexity of CNNs has been also addressed from the hardware-point of view by introducing Graphics Processing Units (GPUs) [20] [21] or custom hardware (Field-Programmable Gate Array -FPGA- or ad-hoc hardware) [22] [17].”
BackgroundEnergy-Efficient Architecture for FPGA-based Deep Convolutional Neural Networks with Binary Weights
Duan, Y., Li, S., Zhang, R., Wang, Q., Chen, J., & Sobelman, G. E. (2018) · IEEE International Conference on Digital Signal Processing (DSP) · doi:10.1109/icdsp.2018.8631596 · Cited by 15
Transaction-level Model Simulator for Communication-Limited AcceleratorsPreprint
Kim, S., Wang, J., Seo, Y., Lee, S., Park, Y., Park, S., & Park, C. S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.14897 · Cited by 15
“In this case, these tile sizes are no longer layer-dependent since the unroll factors are mostly layer-independent, i.e., uniform across different convolution layers [2], [4].”
BackgroundMethodInfluentialAn efficient loop tiling framework for convolutional neural network inference accelerators
Huang, H., Hu, X., Li, X., & Xiong, X. (2021) · IET Circuits Devices & Systems · doi:10.1049/cds2.12091 · Cited by 14
AdaPyramid: Adaptive Pyramid for Accelerating High-Resolution Object Detection on Edge Devices
Shi, X., Zhang, S., Wu, J., Chen, N., Cheng, K., Liang, Y., & Lu, S. (2023) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2023.3343448 · Cited by 14
“For example, various accelerators such as GPU, FPGA [22] and ASIC [23] are developed with specialized drivers, boosting the computing resources on edge devices.”
BackgroundJung, K., Woo, J., & Mukhopadhyay, S. (2023) · IEEE Access · doi:10.1109/access.2023.3344175 · Cited by 14
“Within this scope, FPGA and ASIC-based accelerators for Deep Neural Networks (DNNs) stand out for their impressive energy efficiency and lower power consumption compared to CPUs and GPUs [23], [24], [25], [26], [27], [28].”
BackgroundLevelST: Stream-based Accelerator for Sparse Triangular Solver
He, Z., Song, L., Lucas, R. F., & Cong, J. (2024) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3626202.3637568 · Cited by 14
“For solving a linear system, performance engineers usually reorder the sparse matrix first, then decompose the matrix into lower and upper triangular systems before solving them separately.”
BackgroundSelf-citationEnergy-Efficient Implementation of YOLOv8, Instance Segmentation, and Pose Detection on RISC-V SoC
Wang, H., Li, D., & Isshiki, T. (2024) · IEEE Access · doi:10.1109/access.2024.3397536 · Cited by 14
“Numerous established loop optimization methodologies, including but not limited to loop unrolling, tiling, and interchange [31], prove to be highly effective in mitigating these challenges.”
BackgroundLv, Z., Lu, Z., Xia, K., Zhang, L., Zuo, H., Xu, Y.-W., & Liu, K. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.125041 · Cited by 14
FPGA-accelerated hybrid CNN-LSTM system for efficient EEG-based drowsiness recognition
Yanamala, R. M. R., & Muralidhar, P. (2025) · The Journal of Supercomputing · doi:10.1007/s11227-025-06947-y · Cited by 14
FPGA-accelerated mode decomposition for multimode fiber-based communication
Zhang, Q., Zhang, Y., & Czarske, J. (2025) · Light Advanced Manufacturing · doi:10.37188/lam.2025.031 · Cited by 14
“This design, based on the approach described in [31], implements efficient hardware acceleration by processing convolutional layers of a CNN sequentially.”
Toward transforming space exploration with artificial intelligence neuromorphic computing
Yu, A., Woo, S., & Ahn, H. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.111055 · Cited by 14
Advancements in neural network acceleration: a comprehensive review
Liu, Y.-H., Chang, Y., & Ma, Y. (2025) · ICT Express · doi:10.1016/j.icte.2025.10.015 · Cited by 14
CXL-SpecKV: A Disaggregated FPGA Speculative KV-Cache for Datacenter LLM Serving
Liu, D., & Yu, Y. (2026) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3748173.3779188 · Cited by 14
FPGA-Accelerated Hadoop Cluster for Deep Learning Computations
Alhamali, A., Salha, N., Morcel, R., Ezzeddine, M., Hamdan, O., Akkary, H. H., & Hajj, H. M. (2015) · IEEE ... International Conference on Data Mining workshops · doi:10.1109/icdmw.2015.148 · Cited by 14
“The implementation is profiled for performance bottlenecks using VTune, an Intel profiling tool, and the convolution is found to consume 90.15% of the total computation time.”
MethodInfluentialBridge the Gap between Neural Networks and Neuromorphic Hardware with a Neural Network Compiler
Ji, Y., Zhang, Y., Chen, W., & Xie, Y. (2018) · ACM SIGPLAN Notices · doi:10.1145/3296957.3173205 · Cited by 14
Efficient Design of Pruned Convolutional Neural Networks on FPGA
Véstias, M. P. (2020) · Journal of Signal Processing Systems · doi:10.1007/s11265-020-01606-2 · Cited by 14
“The former uses an architecture similar to that proposed in [17].”
MethodA Deep Neural Network Accelerator using Residue Arithmetic in a Hybrid Optoelectronic System
Peng, J., Alkabani, Y., Puri, K., Ma, X., Sorger, V. J., & El‐Ghazawi, T. (2022) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3550273 · Cited by 14
“Multiple works provide diferent strategies to address these challenges in both conventional [10, 11, 21, 27, 32, 67, 70] and emerging technologies [3, 33, 51, 53, 56, 68].”
BackgroundTensor slicing and optimization for multicore NPUs
de Sousa, R. T., Pereira, M. M., Kwon, Y., Kim, T., Jung, N., Kim, C.-S., Frank, M., & Araujo, G. C. S. D. (2022) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2022.12.008 · Cited by 14
“To improve data reuse, some works use polyhedral-based optimization techniques [19], [38].”
MethodHardware Architecture Design for Hand Gesture Recognition System on FPGA
Tsai, T., Ho, Y.-C., & Chi, P. (2023) · IEEE Access · doi:10.1109/access.2023.3277857 · Cited by 14
“Therefore [19] proposed an analysis design using the roofline model, and used various optimization techniques such as cycle scheduling and pacing.”
MethodReconfigurable CNN Accelerator Embedded in Instruction Extended RISC-V Core
Wang, H., Li, D., & Isshiki, T. (2023) · International Conference on Electronics Technology (ICET) · doi:10.1109/icet58434.2023.10211513 · Cited by 14
“These problems can be handled by the existing loop optimization techniques, such as loop unrolling, tiling, and interchange [8].”
BackgroundVenus: A Versatile Deep Neural Network Accelerator Architecture Design for Multiple Applications
Yang, J., Zheng, H., & Louri, A. (2023) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac56929.2023.10247897 · Cited by 14
“The reuse FIFO acts as a double buffer [12], which can support inter-PE communication.”
BackgroundAccelerated Artificial Neural Networks on FPGA for Fault Detection in Automotive Systems
Shreejith, S., Anshuman, B., & Fahmy, S. A. (2016) · Design, Automation, and Test in Europe · doi:10.3850/9783981537079_0337 · Cited by 14
“FPGAs represent an ideal platform for accelerating ANN-based systems because they enable large scale parallelism while also supporting high throughput floating point computations [3], [4], [9].”
BackgroundFASTCF: FPGA-based Accelerator for STochastic-Gradient-Descent-based Collaborative Filtering
Zhou, S., Kannan, R., Min, Y., & Prasanna, V. K. (2018) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3174243.3174252 · Cited by 14
Polymorphic Accelerators for Deep Neural Networks
Mazreah, A. A., & Chen, L. (2021) · IEEE Transactions on Computers · doi:10.1109/tc.2020.3048624 · Cited by 14
A Framework for Modeling, Optimizing, and Implementing DNNs on FPGA Using HLS
Shahshahani, M., Khabbazan, B., Sabri, M., & Bhatia, D. K. (2020) · IEEE Dallas Circuits and Systems Conference (DCAS) · doi:10.1109/dcas51144.2020.9330667 · Cited by 14
“Each layer weights can be stored in the on-chip FPGA memory to reduce the data accesses[2, 3].”
BackgroundMapping Multiple LSTM models on FPGAs
Ribes, S., Trancoso, P., Sourdis, I., & Bouganis, C.-S. (2020) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/icfpt51103.2020.00010 · Cited by 14
“For estimating the execution time of our FPGA accelerator we derive a roofline model for calculating the attainable performance of the possible designs [14], [15].”
MethodHybrid CNN-SVM Inference Accelerator on FPGA Using HLS
Liu, B., Zhou, Y., Feng, L., Fu, H., & Fu, P. (2022) · Electronics · doi:10.3390/electronics11142208 · Cited by 14
On the Automation of High Level Synthesis of Convolutional Neural Networks
Sozzo, E. D., Solazzo, A., Miele, A., & Santambrogio, M. D. (2016) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw.2016.153 · Cited by 14
“For this reason, in the last years, different works have proposed to accelerate CNNs on Graphic Processing Unit (GPU), Field Programmable Gate Array (FPGA) and even Application Specific Integrated Circuit (ASIC) [7, 8, 9] in order to achieve high performance in terms of throughput or turnaround time.”
BackgroundMethodInfluentialSTT-RAM Buffer Design for Precision-Tunable General-Purpose Neural Network Accelerator
Song, L., Wang, Y., Han, Y., Li, H., Cheng, Y., & Li, X. (2017) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2016.2644279 · Cited by 14
“learning accelerators [13], [15], [16], which consists of buffer, computational block (neural Processing Unit, PE), and logic control unit.”
BackgroundMethodThe Design of Lightweight and Multi Parallel CNN Accelerator Based on FPGA
Li, Z., Lu-yuan, W., Yu, J., Bo-wen, C., & Liang, H. (2019) · IEEE Joint International Information Technology and Artificial Intelligence Conference (ITAIC) · doi:10.1109/itaic.2019.8785800 · Cited by 14
Chen, J.-L., Ai, P., Wang, D., Wang, H., Fang, N., Xu, D., Gong, Q., & Yang, Y.-K. (2020) · Nuclear Science and Techniques · doi:10.1007/s41365-020-00756-z · Cited by 14
“Compared with general neural network accelerators [14, 24], this design uses less resources.”
Bus Width Aware Off-Chip Memory Access Minimization for CNN Accelerators
Tewari, S., Kumar, A., & Paul, K. (2020) · IEEE Computer Society Annual Symposium on VLSI · doi:10.1109/isvlsi49217.2020.00051 · Cited by 14
“The off-chip memory accesses of a layer depends significantly on the tile dimensions [1], [12].”
BackgroundMethodInfluentialREQ-YOLO: A Resource-Aware, Efficient Quantization Framework for Object Detection on FPGAsPreprint
Ding, C., Wang, S., Liu, N., Xu, K., Wang, Y., & Liang, Y. E. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1909.13396 · Cited by 15
“The second one is efficient hardware implementations, including FPGAs and ASICs [1, 7, 31, 32, 36, 49, 50, 53, 54].”
MethodFeCaffe: FPGA-enabled Caffe with OpenCL for Deep Learning Training and Inference on Intel Stratix 10
He, K., Liu, B., Zhang, Y., Ling, A., & Gu, D. (2020) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3373087.3375389 · Cited by 14
Resource-Aware Saliency-Guided Differentiable Pruning for Deep Neural Networks
Kallakuri, U., Humes, E., & Mohsenin, T. (2024) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3649476.3658699 · Cited by 13
On-Edge High-Throughput Collaborative Inference for Real-Time Video Analytics
Wang, X., Shen, M., & Yang, K. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3424235 · Cited by 13
“Moreover, some model acclerators, such as FPGA [40] and ASIC [41] are designed to achieve high-throughput and low-delay DNN model inference.”
MethodOn-Device Deep Learning: Survey on Techniques Improving Energy Efficiency of DNNs
Boumendil, A., Bechkit, W., & Benatchba, K. (2024) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2024.3430028 · Cited by 13
“The designing techniques of such platforms are outside the scope of this article, but examples of the latter can be found in [31], [134], [135], [136], [137], [138], [139], [140], [141], [142], and [143].”
Recent advances in Machine Learning based Advanced Driver Assistance System applications
Tatar, G., Bayar, S., Çiçek, İ., & Niar, S. (2024) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2024.105101 · Cited by 14
A Scalable, Efficient, and Robust Dynamic Memory Management Library for HLS-based FPGAs
Wang, Q., Zheng, L., An, Z., Xiong, S., Wang, R., Huang, Y., Yao, P., Liao, X., Jin, H., & Xue, J. (2024) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro61859.2024.00040 · Cited by 13
“These HLS tools have gained widespread adoption in numerous application developments [8], [10], [60].”
Sun, X., Zhang, Y., Wang, Q., Zou, X., Liu, Y., Zeng, Z., & Zhuang, H. (2025) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2025.105042 · Cited by 13
MetaML-Pro: Cross-Stage Design Flow Automation for Efficient Deep Learning Acceleration
Que, Z., Coutinho, J. G. F., Guo, C., Fan, H., & Luk, W. W. (2026) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3795794 · Cited by 13
FPGA-based CNN accelerator using Convolutional Processing Element to reduce idle states
Dehnavi, M., Ghasemi, A., & Alizadeh, B. (2025) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2025.103468 · Cited by 13
Liu, L., Su, B., Jiang, J., Wu, G., Guo, C., Xu, C., & Yang, H. F. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.02403 · Cited by 13
“Neverthe-less, these approaches [4] often impose substantial hardware overhead and exhibit limited adaptability, particularly when deployed on resource-constrained edge platforms [5].”
A Quantized Neural Network Library for Proper Implementation of Hardware Emulation
Kiyama, M., Nakahara, Y., Amagasaki, M., & Iida, M. (2019) · 2019 Seventh International Symposium on Computing and Networking Workshops (CANDARW) · doi:10.1109/candarw.2019.00032 · Cited by 5
“Many researches [15]–[17] have been studied for effective implementation on embedded systems.”
Background
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.