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 601–700 of 1,002
Software-Defined Design Space Exploration for an Efficient DNN Accelerator Architecture
Yu, Y., Li, Y., Che, S., Jha, N. K., & Zhang, W. (2020) · IEEE Transactions on Computers · doi:10.1109/tc.2020.2983694 · Cited by 31
“change are often used in recent DNN accelerators [29], [30].”
MethodA FPGA-based Accelerator of Convolutional Neural Network for Face Feature Extraction
Ding, R., Su, G., Bai, G., Xu, W., Su, N., & Wu, X. (2019) · IEEE International Conference on Electron Devices and Solid-State Circuits (EDSSC) · doi:10.1109/edssc.2019.8754067 · Cited by 31
“Using this method can achieve optimal hardware resource utilization according to each CONV layers characteristics [7].”
MethodFPGA-Based Training Accelerator Utilizing Sparseness of Convolutional Neural Network
Nakahara, H., Sada, Y., Shimoda, M., Sayama, K., Jinguji, A., & Sato, S. (2019) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2019.00036 · Cited by 31
“However, these FC layers require many parameters, and this causes an overfitting problems [32].”
BackgroundMethodFast-BCNN: Massive Neuron Skipping in Bayesian Convolutional Neural Networks
Wan, Q., & Fu, X. (2020) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro50266.2020.00030 · Cited by 31
“Current parallelism strategies can be categorized into three types [20]: Synapse Parallelism [21], Neuron Parallelism [22] and Feature map Parallism [23].”
MethodReconfigurable Hardware Accelerators: Opportunities, Trends, and ChallengesPreprint
Wang, C., Lou, W., Gong, L., Jin, L., Tan, L., Hu, Y., Li, X., & Zhou, X. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1712.04771 · Cited by 32
“Reference [57] uses Roofline model analysis design to optimize FPGAbased CNN accelerators.”
MethodAccuracy vs. Efficiency: Achieving Both through FPGA-Implementation Aware Neural Architecture SearchPreprint
Jiang, W., Zhang, X., Sha, E. H. ‐., Yang, L., Zhuge, Q., Shi, Y., & Hu, J. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1901.11211 · Cited by 31
“With such vision, a series of works for implementing neural networks on FPGAs have been carried out [3, 4, 8, 9, 11, 13, 15].”
BackgroundMethodInfluentialNAS4RRAM: neural network architecture search for inference on RRAM-based accelerators
Yuan, Z., Liu, J., Li, X., Yan, L., Chen, H., Wu, B., Yang, Y., & Sun, G. (2021) · Science China Information Sciences · doi:10.1007/s11432-020-3245-7 · Cited by 30
Self-citationA Survey of Artificial Neural Network Computing Systems
Foukalas, F. T. (2024) · Cognitive Computation · doi:10.1007/s12559-024-10383-0 · Cited by 30
FPGA-QNN: Quantized Neural Network Hardware Acceleration on FPGAs
Tasci, M., İstanbullu, A., Tümen, V., & Koşunalp, S. (2025) · Applied Sciences · doi:10.3390/app15020688 · Cited by 31
When FPGA-Accelerator Meets Stream Data Processing in the Edge
Wu, S., Hu, D., Ibrahim, S., Jin, H., Xiao, J., Chen, F., & Liu, H. (2019) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs.2019.00180 · Cited by 30
“…are widely deployed accelerators in data centers (e.g., Microsoft have deployed FPGAs on 1632 servers to accelerate the Bing web search engine [19]) and have been successfully used to accelerate data mining [20], graph processing [21], deep learning [22, 23], and stream data processing [24–26].”
MethodNAIS: Neural Architecture and Implementation Search and its Applications in Autonomous Driving
Hao, C., Chen, Y., Liu, X., Sarwari, A., Sew, D., Dhar, A., Wu, B., Fu, D., Xiong, J., Hwu, W., et al. (2019) · doi:10.1109/iccad45719.2019.8942055 · Cited by 30
Di, X., Yang, H., Jia, Y., Huang, Z., & Mao, N. (2020) · Electronics · doi:10.3390/electronics9020286 · Cited by 30
“Actually, neither the work of Zhang [18] nor GNA [17] implements real-life GAN models.”
MethodInfluentialFast Algorithms for Quaternion-Valued Convolutional Neural Networks
Cariow, A., & Cariowa, G. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.2979682 · Cited by 30
“In CNN, convolutional layers are the most computationally intensive since, in a typical implementation, they occupied more than 90% of the CNN execution time [4].”
BackgroundMethodCODEBench: A Neural Architecture and Hardware Accelerator Co-Design Framework
Tuli, S., Li, C., Sharma, R., & Jha, N. K. (2023) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3575798 · Cited by 30
“Recent works factor this in with a memory bandwidth parameter [51, 85].”
BackgroundPLACID: A Platform for FPGA-Based Accelerator Creation for DCNNs
Motamedi, M., Gysel, P., & Ghiasi, S. (2017) · ACM Transactions on Multimedia Computing, Communications, and Applications · doi:10.1145/3131289 · Cited by 28
“Zhang et al. [30] proposed using tiling over row and columns of IFMs.”
BackgroundMethodResultInfluentialFast generation of high throughput customized deep learning accelerators on FPGAs
Zeng, H., Zhang, C., & Prasanna, V. K. (2017) · International Conference on ReConFigurable Computing and FPGAs (ReConFig) · doi:10.1109/reconfig.2017.8279792 · Cited by 30
“Apart from GPU, FPGA has also become an attractive platform for CNN acceleration [5].”
BackgroundSelf-citationDendaluce, M., Cosco, F., Novickis, R., Perez, J., & Garay, V. (2019) · Electronics · doi:10.3390/electronics8020250 · Cited by 31
Transmuter: Bridging the Efficiency Gap using Memory and Dataflow Reconfiguration
Pal, S., Feng, S., Park, D.-H., Kim, S., Amarnath, A., Yang, C.-S., He, X. J., Beaumont, J., May, K., Xiong, Y., et al. (2020) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1145/3410463.3414627 · Cited by 30
Spatio-Temporal Optimization of Deep Neural Networks for Reconfigurable FPGA SoCs
Seyoum, B. B., Pagani, M., Biondi, A., Balleri, S., & Buttazzo, G. (2020) · IEEE Transactions on Computers · doi:10.1109/tc.2020.3033730 · Cited by 30
“Most relevant to us, Zhang et al. [22] leveraged the roofline model for optimizing HLS-based implementations of DNNs, whereas Ma et al. [23] performed a fine-grained characterization of loop optimization techniques for implementing convolutional layers.”
MethodSERF: Efficient Scheduling for Fast Deep Neural Network Serving via Judicious Parallelism
Yan, F., Ruwase, O., He, Y., & Smirni, E. (2016) · International Conference for High Performance Computing, Networking, Storage and Analysis (SC) · doi:10.1109/sc.2016.25 · Cited by 30
“Several recent work on DNN serving investigate hardware acceleration using GPUs [44], FPGAs [45], and ASICs [28].”
BackgroundUniWiG: Unified Winograd-GEMM Architecture for Accelerating CNN on FPGAs
Kala, S., Mathew, J., Jose, B. R., & Nalesh, S. (2019) · International Conference on VLSI Design · doi:10.1109/vlsid.2019.00055 · Cited by 29
“Various implementations of Winograd Filtering based convolution have been proposed during the last three years [1], [6]–[8], [12], [13].”
MethodA Resources-Efficient Configurable Accelerator for Deep Convolutional Neural Networks
Hu, X., Zeng, Y., Li, Z., Zheng, X. W., Cai, S., & Xiong, X. (2019) · IEEE Access · doi:10.1109/access.2019.2919527 · Cited by 29
“Zhang et al. [6] and Motamedi et al. [14] identify the best system performance with roofline model based on the peak performance and off-chip memory traffic provided by the hardware platform.”
BackgroundMethodInfluentialGhodrati, S., Sharma, H., Kinzer, S., Yazdanbakhsh, A., Park, J., Kim, N. S., Burger, D., & Esmaeilzadeh, H. (2020) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1145/3410463.3414634 · Cited by 29
“With the diminishing benefits from general-purpose processors [6–9], there is an explosion of digital accelerators for DNNs [10–31].”
BackgroundImplementation of Data-optimized FPGA-based Accelerator for Convolutional Neural Network
Cho, M., & Kim, Y. (2020) · International Conference on Electronics, Information, and Communication (ICEIC) · doi:10.1109/iceic49074.2020.9050993 · Cited by 29
“When designing hardware designs for FPGA platform, High-level Synthesis (HLS) can be used to efficiently implement hardware accelerators [4][5].”
MethodWooKong: A Ubiquitous Accelerator for Recommendation Algorithms with Custom Instruction Sets on FPGA
Wang, C., Gong, L., Ma, X., Li, X., & Zhou, X. (2020) · IEEE Transactions on Computers · doi:10.1109/tc.2020.2988209 · Cited by 29
“For example, Zhang et al. [25] explores the bandwidth for the parameters facing the limitation of an FPGA chip.”
BackgroundAdaPrune: An Accelerator-Aware Pruning Technique for Sustainable CNN Accelerators
Li, J., & Louri, A. (2021) · IEEE Transactions on Sustainable Computing · doi:10.1109/tsusc.2021.3060690 · Cited by 32
FARNN: FPGA-GPU Hybrid Acceleration Platform for Recurrent Neural Networks
Cho, H., Lee, J., & Lee, J. (2021) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2021.3124125 · Cited by 29
“1: Structure of an RNN. [36], [37], [38], [39], [40].”
Duan, X., Chen, Y., Li, M., Rong, Y., Xie, R., & Han, J. S. (2023) · IEEE Transactions on Biomedical Circuits and Systems · doi:10.1109/tbcas.2023.3261060 · Cited by 29
“Previous study [38] uses the tiling technique to slice the convolution computation of a single layer, matching limited on-chip memory.”
MethodLiu, Z., Liu, Q., Yan, S., & Cheung, R. C. C. (2023) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3615661 · Cited by 29
“According to the dimensions of the convolution regarded as 6-level loop [29] in CNNs, the proposed accelerator adopts five parallelization strategies, including PK (parallelism on kernel dimension), PV (parallelism on vector dimension), PC (parallelism on input channel dimension), PF (parallelism…”
MethodEnériz, D., Rodríguez-Almeida, A. J., Fabelo, H. A., Ortega, S., Balea-Fernandez, F., Callicó, G. M., Medrano, N. J., & Calvo, B. (2024) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2024.3392271 · Cited by 29
“Thus, one of the more potential ways to accelerate this model is to perform loop unrolling and pipelining [31], which are the basic directives used in any loop optimization process.”
Implementation and Optimization of the Accelerator Based on FPGA Hardware for LSTM Network
Zhang, Y., Wang, C., Gong, L., Lu, Y., Sun, F., Xu, C., Li, X., & Zhou, X. (2017) · IEEE International Symposium on Parallel and Distributed Processing with Applications (ISPA) · doi:10.1109/ispa/iucc.2017.00098 · Cited by 29
“There are a lot of research works on the use of FPGA for various types of neural networks to achieve acceleration [14][15][16].”
BackgroundExploiting Parallelism for CNN Applications on 3D Stacked Processing-In-Memory Architecture
Wang, Y., Chen, W., Yang, J., & Li, T. (2018) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2018.2868062 · Cited by 29
“FPGA implementations [20], [21], [22], [23] to provide computationally and power efficient systems [24].”
A portable, automatic data qantizer for deep neural networks
Oh, Y. H., Quan, Q., Kim, D.-Y., Kim, S., Heo, J., Jung, S., Jang, J.-Y., & Lee, J. W. (2018) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1145/3243176.3243180 · Cited by 29
“[59] to achieve competitive performance to theirs.”
MulNet: A Flexible CNN Processor With Higher Resource Utilization Efficiency for Constrained Devices
Hailesellasie, M., & Hasan, S. R. (2019) · IEEE Access · doi:10.1109/access.2019.2907865 · Cited by 29
“A few of the highlights of these approaches include, parameter reduction [11], [26], binary weight quantization [20], [21], [24], optimization for power [12], memory bandwidth optimization [15], [16], [23], pipelining, parallelism and batch-based processing [10], [13], [25], computation load reduction [14], and datafl…”
BackgroundA Primer on Deep Learning Architectures and Applications in Speech Processing
Ogunfunmi, T., Ramachandran, R. P., Togneri, R. B., Zhao, Y., & Xia, X. (2019) · Circuits Systems and Signal Processing · doi:10.1007/s00034-019-01157-3 · Cited by 29
“There are acceleration solutions for GPU [15], CPU [2,52,93], FPGA and ASIC platforms [3,11,13,14,35,83,104].”
BackgroundMethodHigh Performance Accelerator for CNN Applications
Kyriakos, A., Kitsakis, V., Louropoulos, A., Papatheofanous, E. A., Patronas, I., & Reisis, D. I. (2019) · International Workshop on Power and Timing Modeling, Optimization and Simulation (PATMOS) · doi:10.1109/patmos.2019.8862166 · Cited by 29
“Existing FPGA accelerators are implemented with automated software tools like HLS [4], [5], [6], that offer”
MethodEfficient algorithms for task mapping on heterogeneous CPU/GPU platforms for fast completion time
Li, Z., Zhang, Y., Ding, A., Zhou, H., & Liu, C. (2020) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2020.101936 · Cited by 29
Scheduling Computation Graphs of Deep Learning Models on Manycore CPUsPreprint
Tang, L., Wang, Y., Willke, T. L., & Li, K. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1807.09667 · Cited by 29
A Survey on Deep Learning Methods in Image Analytics
Vishwakarma, P. K., & Jain, N. (2022) · Lecture notes on data engineering and communications technologies · doi:10.1007/978-981-16-6289-8_52 · Cited by 28
Charles Rajesh Kumar., J., Vinod Kumar, D., Baskar, D., Mary Arunsi., B., Jenova, R., & Majid, M. A. (2019) · Procedia Computer Science · doi:10.1016/j.procs.2019.12.145 · Cited by 28
An FPGA Implementation of Stochastic Computing-Based LSTM
Maor, G., Zeng, X., Wang, Z., & Hu, Y. (2019) · IEEE International Conference on Computer Design: VLSI in Computers and Processors (ICCD) · doi:10.1109/iccd46524.2019.00014 · Cited by 28
“[21], [22] used GPGPU to accelerate the CNN implementation; [23], [24] explored the optimization on CNN using FPGA [12].”
MethodDesign Space Exploration for YOLO Neural Network Accelerator
Huang, H., Liu, Z., Chen, T., Hu, X., Zhang, Q., & Xiong, X. (2020) · Electronics · doi:10.3390/electronics9111921 · Cited by 28
“The authors of [20] presented a 32-bit floating-point design based on ImageNet, which achieves 61.”
BackgroundMethodInfluentialBatch Normalization Processor Design for Convolution Neural Network Training and Inference
Ting, Y.-S., Teng, Y.-F., & Chiueh, T. (2021) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas51556.2021.9401434 · Cited by 28
Zhang, X., Hao, C., Zhou, P., Jones, A. K., & Hu, J. (2022) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3489517.3530509 · Cited by 28
“The data transfer overhead, however, is not considered.”
Horizontally Distributed Inference of Deep Neural Networks for AI-Enabled IoT
Rodríguez, I., Campos, C., & Fdez‐Riverola, F. (2023) · Sensors · doi:10.3390/s23041911 · Cited by 28
“AI-specific capabilities of novel hardware solutions —including embedded devices and accessories for accelerating ML tasks, such as applicationspecific integrated circuits (ASICs) [38,39], GPUs [40,41], and FPGAs [42,43]—have emerged as the infrastructure cornerstone of the recent AI trend known as on-device ML, under…”
A high-efficiency runtime reconfigurable IP for CNN acceleration on a mid-range all-programmable SoC
Meloni, P., Deriu, G., Conti, F., Loi, I., Raffo, L., & Benini, L. (2016) · International Conference on ReConFigurable Computing and FPGAs (ReConFig) · doi:10.1109/reconfig.2016.7857144 · Cited by 28
“Zhang et al. [16] propose a CNN acceleration resident on FPGA fabric that by using optimal tiling of the convolutional layers can avoid much of the overhead to move partial results.”
BackgroundMethodA 200MHZ 202.4GFLOPS@10.8W VGG16 accelerator in Xilinx VX690T
Mei, C., Liu, Z., Niu, Y., Ji, X., Zhou, W., & Wang, D. (2017) · IEEE Global Conference on Signal and Information Processing · doi:10.1109/globalsip.2017.8309067 · Cited by 28
“Comparing to HLS design in previous works [3,4], resource utilization and power consumption of our RTL oriented architecture are more efficient.”
MethodBlock convolution: Towards memory-efficient inference of large-scale CNNs on FPGA
Li, G., Li, F., Zhao, T., & Cheng, J. (2018) · Design, Automation, and Test in Europe · doi:10.23919/date.2018.8342188 · Cited by 28
ShmCaffe: A Distributed Deep Learning Platform with Shared Memory Buffer for HPC Architecture
Ahn, S., Kim, J., Lim, E., Choi, W., Mohaisen, A., & Kang, S. (2018) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs.2018.00111 · Cited by 28
“In addition, some hardware accelerators, including FPGAs and Intel Xeon Phi processors, as well as GPUs, can be used for DNN training [22].”
MethodGeneSys: Enabling Continuous Learning through Neural Network Evolution in Hardware
Samajdar, A., Mannan, P., Garg, K., & Krishna, T. (2018) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro.2018.00074 · Cited by 29
“Hardware acceleration of neural networks is a hot research topic with a lot of architecture choices [49], [50], [51], [52], [17], [53], [54], [55], [56] and silicon implementations [25], [26], [27], [28].”
BackgroundDerin Öğrenme Araştırma Alanlarının Literatür Taraması
Yapıcı, M. M., Tekerek, A., & Topaloğlu, N. (2019) · Gazi Journal of Engineering Sciences · doi:10.30855/gmbd.2019.03.01 · Cited by 29
Programmatic Control of a Compiler for Generating High-performance Spatial HardwarePreprint
Rong, H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1711.07606 · Cited by 28
“How to come up with such a design is beyond the scope of the paper, but a general principle is to fully exploit memory bandwidth and maximize computation to communication ratio [17~20].”
BackgroundVarGNet: Variable Group Convolutional Neural Network for Efficient Embedded ComputingPreprint
Zhang, Q., Li, J., Meng, Y., Song, L., Zhou, H., Li, Z., Meng, W., Zhang, X., & Wang, G. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1907.05653 · Cited by 28
“To accelerate neural networks, FPGAs [13, 50, 17, 31] and ASIC designs [7, 36, 23, 29, 19] have been widely studied.”
BackgroundMethodMarvel: A Data-centric Compiler for DNN Operators on Spatial AcceleratorsPreprint
Chatarasi, P., Kwon, H., Raina, N., Malik, S., Haridas, V., Parashar, A., Pellauer, M., Krishna, T., & Sarkar, V. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.07752 · Cited by 28
“Level-1 tiling is ignored in most of the prior accelerator designs [27, 48, 49, 49], because of the absence of a private buffer inside a PE in those accelerators.”
BackgroundMethodInfluentialBoosting Convolutional Neural Networks Performance Based on FPGA Accelerator
Al-Shamma, O., Fadhel, M. A., Hameed, R. A., Alzubaidi, L., & Zhang, J. (2020) · Advances in Intelligent Systems and Computing · doi:10.1007/978-3-030-16657-1_47 · Cited by 27
“Convolutional Neural Networks (CNNs) [ 1 ] is one of these techniques that become the actual standard technique due to its providing high near-human accuracy in several speech recognition [2] and machine vision (e.g., segmentation [3], detection [4], and classification [5]) applications.”
BackgroundSabogal, S., George, A. D., & Crum, G. (2019) · IEEE Space Computing Conference (SCC) · doi:10.1109/spacecomp.2019.00010 · Cited by 27
“In [19], a roofline model was proposed to explore the design space for possible solutions for a CNN architecture on an FPGA platform, subject to computational resource and memory bandwidth constraints, using loop optimization techniques such as loop unrolling, tiling, and interchange.”
BackgroundMethodDNNARA: A Deep Neural Network Accelerator using Residue Arithmetic and Integrated Photonics
Peng, J., Alkabani, Y., Sun, S., Sorger, V. J., & El‐Ghazawi, T. (2020) · Proceedings of the International Conference on Parallel Processing · doi:10.1145/3404397.3404467 · Cited by 27
“They are all implemented on FPGAs.”
MethodAI accelerator on IBM Telum processor: industrial product
Lichtenau, C., Buyuktosunoglu, A., Bertran, R., Figuli, P., Jacobi, C., Papandreou, N., Pozidis, H., Saporito, A., Sica, A., & Tzortzatos, E. (2022) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3470496.3533042 · Cited by 27
“…including commercially available ones [25, 26, 55], a wide range of accelerators from ASIC, FPGA, bit-serial, in-memory designs to large scale systems have been proposed and implemented [5, 7, 19, 20, 22–24, 31, 32, 35, 36, 38, 39, 55, 56, 58, 60– 63, 70, 71, 77–79, 82, 83, 87, 91, 100, 103, 104].”
Stateful Neural Networks for Intermittent Systems
Yen, C.-H., Mendis, H. R., Kuo, T., & Hsiu, P.-C. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3197513 · Cited by 27
“Due to the limited VM capacity of embedded devices, IFMs, OFMs, and model parameters are stored in NVM and are processed as tiles in VM [11], [32], where VM has lower data access energy and latency costs than NVM.”
MethodSystematic Literature Review on Cost-Efficient Deep Learning
Klemetti, A., Raatikainen, M., Myllyaho, L., Mikkonen, T., & Nurminen, J. K. (2023) · IEEE Access · doi:10.1109/access.2023.3275431 · Cited by 27
End-to-end acceleration of the YOLO object detection framework on FPGA-only devices
Zhang, D., Wang, A., Mo, R., & Wang, D. (2023) · Neural Computing and Applications · doi:10.1007/s00521-023-09078-8 · Cited by 27
“The bandwidth roof measures the throughput that the memory system can support [25].”
A Threshold Neuron Pruning for a Binarized Deep Neural Network on an FPGA
Fujii, T., Sato, S., & Nakahara, H. (2018) · IEICE Transactions on Information and Systems · doi:10.1587/transinf.2017rcp0013 · Cited by 27
Li, H., Li, G., Ji, X., & Shi, L. (2018) · Neurocomputing · doi:10.1016/j.neucom.2018.02.019 · Cited by 27
CNN-on-AWS: Efficient Allocation of Multikernel Applications on Multi-FPGA Platforms
Shan, J., Lazarescu, M. T., Cortadella, J., Lavagno, L., & Casu, M. R. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.2994256 · Cited by 27
“The CUs are optimized using loop tiling and permutation of nested loops to reduce data dependencies and increase parallelism [2].”
MethodEfficient Methods for Mapping Neural Machine Translator on FPGAs
Li, Q., Zhang, X., Xiong, J., Hwu, W., & Chen, D. (2020) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2020.3047371 · Cited by 27
“To accelerate the process of convolutional layers, authors in [6] propose a FPGA-based accelerator by applying loop tiling, unrolling, pipelining and data reuse under memory bandwidth and resource constraints.”
BackgroundMethodMartínez‐Rodrigo, A., García-Martínez, B., Herráiz, Á. H., & Alcaraz, R. (2021) · Sensors · doi:10.3390/s21093050 · Cited by 28
“In this respect, some researchers have already successfully tested a variety of hardware implementation methods for different CNN-based structures, mostly based on field-programmable gate arrays (FPGA) architectures [65,66].”
IECA: An In-Execution Configuration CNN Accelerator With 30.55 GOPS/mm² Area Efficiency
Huang, B., Huan, Y., Chu, H., Xu, J., Liu, L., Zheng, L., & Zou, Z. (2021) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3108762 · Cited by 27
Towards Memory-Efficient Allocation of CNNs on Processing-in-Memory Architecture
Wang, Y., Chen, W., Yang, J., & Li, T. (2018) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2018.2791440 · Cited by 27
“There have been a number of FPGA implementations [1], [12], [13], [14] and ASIC accelerators [15], [16].”
BackgroundZhu, M., Liu, L., Wang, C., & Xie, Y. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1606.06234 · Cited by 27
“The state-of-the-art means for accelerating deep learning algorithms are Field-Programmable Gate Array (FPGA) [2], Application Specific Integrated Circuit (ASIC) [3], and Graphic Processing Unit (GPU) [4].”
BackgroundDNA: Differentiable Network-Accelerator Co-SearchPreprint
Zhang, Y., Fu, Y., Jiang, W., Li, C., You, H., Li, M., Chandra, V., & Lin, Y. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.14778 · Cited by 26
“In this work, we consider three NoC options following the common practice, as inspired by SOTA accelerators (Chen et al., 2016; Zhang et al., 2015; Zhao et al., 2020): • parallelizing the computation over the output partial sums, where the dimensions of output channels, output rows, and output…”
Method\nSoK: Neural Network Extraction Through Physical Side Channels
Horváth, P., Lauret, D., Liu, Z., & Batina, L. (2024) · Radboud Repository (Radboud University) · Cited by 27
Nagasu, K., Sano, K., Kono, F., & Nakasato, N. (2016) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2016.12.015 · Cited by 26
FastWave: Accelerating Autoregressive Convolutional Neural Networks on FPGA
Hussain, S., Javaheripi, M., Neekhara, P., Kastner, R., & Koushanfar, F. (2019) · 2019 IEEE/ACM International Conference on Computer-Aided Design (ICCAD) · doi:10.1109/iccad45719.2019.8942122 · Cited by 26
“to achieve energy efficient neural network computation [12], [13], [22], [24]–[27].”
BackgroundHigh-Performance Acceleration of 2-D and 3-D CNNs on FPGAs Using Static Block Floating Point
Fan, H., Liu, S., Que, Z., Niu, X., & Luk, W. W. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3116302 · Cited by 27
“Another design methodology is to deploy a single processing engine (PE) for CNN-like operations on the FPGA and accelerate the CNN layer-by-layer sequentially using the same hardware engine [22].”
MethodLeveraging Domain Information for the Efficient Automated Design of Deep Learning Accelerators
Sakhuja, C., Shi, Z., & Lin, C. (2023) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca56546.2023.10071095 · Cited by 26
“Other work restricts the hardware design space either by using hardware templates [17], [68], by limiting the design parameters that are searched [71], or by generating hardware that is highly specialized for the given algorithm [62], [64], [74].”
MethodLi, C., Xu, R., Lv, Y., Zhao, Y., & Jing, W. (2023) · Remote Sensing · doi:10.3390/rs15163975 · Cited by 27
Diffraction‐Driven Parallel Convolution Processing with Integrated Photonics
Huang, Y., Liu, W., Sun, R., Fu, T., Wang, Y., Huang, Z., Yang, S., & Chen, H. (2024) · Laser & Photonics Review · doi:10.1002/lpor.202400972 · Cited by 28
AdaPI: Facilitating DNN Model Adaptivity for Efficient Private Inference in Edge Computing
Zhou, T., Zhao, J., Luo, Y., Xie, X., Wen, W., Ding, C., & Xu, X. (2024) · doi:10.1145/3676536.3676694 · Cited by 28
“The computation part follows tiled architecture implementation [37].”
Stochastic-Based Deep Convolutional Networks with Reconfigurable Logic Fabric
Alawad, M., & Lin, M. (2016) · IEEE Transactions on Multi-Scale Computing Systems · doi:10.1109/tmscs.2016.2601326 · Cited by 26
“Most of the researches have focused on either improving the performance of computing circuit or the global communication with the off-chip memory issue, although a few of them have tried to improve both, such as [9].”
BackgroundFPGA-Based Real-Time Super-Resolution System for Ultra High Definition Videos
He, Z., Huang, H., Jiang, M., Bai, Y., & Luo, G. (2018) · IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM) · doi:10.1109/fccm.2018.00036 · Cited by 26
“They explore parallelism, computing sequences (pipelines), and computation-communication balance by loop optimization techniques like loop unrolling and loop tiling [15]–[18].”
Application Partitioning on FPGA Clusters: Inference over Decision Tree Ensembles
Owaida, M., & Alonso, G. (2018) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2018.00057 · Cited by 26
Live Video Analytics with FPGA-based Smart Cameras
Wang, S., Zhang, C., Shu, Y., & Liu, Y. (2019) · doi:10.1145/3349614.3356027 · Cited by 28
Self-citationApproximate Acceleration for CNN-based Applications on IoT Edge Devices
Godinez, J. C., Hernandez-Araya, D., Shafique, M., & Henkel, J. (2020) · IEEE Latin American Symposium on Circuits and Systems (LASCAS) · doi:10.1109/lascas45839.2020.9069040 · Cited by 26
“…main scenarios have been reported in the literature regarding to custom hardware acceleration for CNNs: • General Purpose Processor and off-the-shelf FPGA: Spe-cific computations of CNNs are moved to hardware blocks on the FPGA [9], [10], and specialized co-processors are deployed on FPGAs [11].”
BackgroundPointNet on FPGA for Real-Time LiDAR Point Cloud Processing
Bai, L., Lyu, Y., Xu, X., & Huang, X. (2020) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas45731.2020.9180841 · Cited by 26
“Newly proposed architectures [9][10] for neural networks take advantage of Single Instruction Multiple Data (SIMD) structure for matrix multiplication.”
BackgroundKhatwani, M., Rashid, H.-A., Paneliya, H., Horton, M., Waytowich, N. R., Hairston, W. D., & Mohsenin, T. (2021) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3427471 · Cited by 26
“ACM Reference Format: Mohit Khatwani, Hasib-Al Rashid, Hirenkumar Paneliya, Mark Horton, Nicholas Waytowich, W. David Hairston, and Tinoosh Mohsenin.”
BackgroundMethodInfluentialCustomizable Computing—From Single Chip to Datacenters
Cong, J., Fang, Z., Huang, M., Wei, P., Wu, D., & Yu, C. H. (2018) · Proceedings of the IEEE · doi:10.1109/jproc.2018.2876372 · Cited by 26
“[83] C.”
Self-citationDeep3: Leveraging Three Levels of Parallelism for Efficient Deep Learning
Rouhani, B. D., Mirhoseini, A., & Koushanfar, F. (2017) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3061639.3062225 · Cited by 25
“(ii) Computer engineers, on the other hand, have developed hardware-accelerated solutions for an efficient domain-customized realization of DL models, e.g., [5, 6].”
BackgroundAn energy-efficient and high-throughput bitwise CNN on sneak-path-free digital ReRAM crossbar
Ni, L., Liu, Z., Song, W., Yang, J. J., Yu, H., Wang, K., & Wang, Y. (2017) · International Symposium on Low Power Electronics and Design (ISLPED) · doi:10.1109/islped.2017.8009177 · Cited by 25
Kim, M., & Seo, J.-S. (2020) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2020.3029235 · Cited by 25
“To efficiently map and perform the convolution loops, three loop optimization techniques [23]–[25] of loop unrolling, loop tiling, and loop interchange are employed to customize the computation and communication patterns of the accelerator with three levels of memory hierarchy.”
MethodHardware acceleration for object detection using YOLOv4 algorithm on Xilinx Zynq platform
Babu, P., & Parthasarathy, E. (2022) · Journal of Real-Time Image Processing · doi:10.1007/s11554-022-01234-y · Cited by 26
“By the reason of a large amount of input data, loop pipelining [26] is used to enhance system throughput in high-level synthesis.”
MethodInfluentialSheng, Y., Yang, J., Wu, Y., Mao, K., Shi, Y., Hu, J., Jiang, W., & Yang, L. (2022) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3489517.3530427 · Cited by 25
Algorithm/Hardware Co-Optimization for Sparsity-Aware SpMM Acceleration of GNNs
Gao, Y., Gong, L., Wang, C., Wang, T., Li, X., & Zhou, X. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3281714 · Cited by 25
“3) Ping-Pong Buffer: Due to the long off-chip access latency, many previous accelerators adopt the decoupling design of the computation module and memory access module [26], [27], [28].”
BackgroundMethodOptimizing FPGA-Based DNN Accelerator With Shared Exponential Floating-Point Format
Zhao, W., Dang, Q., Xia, T., Zhang, J., Zheng, N., & Ren, P. (2023) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2023.3300657 · Cited by 25
“Early works such as Zhang et al. [33] implemented DNN accelerators on FPGAs with regular FP 32 format, which fail to provide the computing power to deal with large neural networks.”
BackgroundHussain, M. (2024) · AI · doi:10.3390/ai5030064 · Cited by 25
“…this issue, consisting of several sub-components; loop reordering for blocking redundant memory access between loops enhancing cache capacity [114], loop unrolling, and pipe-lining improve FPGA resource utilisation [115,116], whilst loop tiling, partitions weights/feature maps for each layer…”
Enhanced Efficiency 3D Convolution Based on Optimal FPGA Accelerator
Wang, H., Shao, M., Liu, Y., & Zhao, W. (2017) · IEEE Access · doi:10.1109/access.2017.2699229 · Cited by 25
“In [11], the authors leveraged all sources of parallelism in CNNs and their work ran 84.2 GFLOPS on a Virtex-7 FPGA.”
BackgroundInfluentialA pipelined and scalable dataflow implementation of convolutional neural networks on FPGA
Bacis, M., Natale, G., Sozzo, E. D., & Santambrogio, M. D. (2017) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw.2017.44 · Cited by 25
“Previous work on FPGA-based implementation of CNNs mostly focus on the acceleration of the convolutional layer [10, 14, 12], being the most compute-intensive layer, or propose solutions that do not fully exploit the suitability of CNNs for a dataflow implementation [15, 16].”
BackgroundA Case of On-Chip Memory Subsystem Design for Low-Power CNN Accelerators
Wang, Y., Li, H., & Li, X. (2017) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2017.2778060 · Cited by 25
“Other coarse-grained programmable architectures can exploit the massive parallelism within the feature maps and convolution kernel through loop unrolling in the array of PEs [5], [7].”
BackgroundCaFPGA: An automatic generation model for CNN accelerator
Xu, J., Liu, Z.-Q., Jiang, J., Dou, Y., & Li, S. (2018) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2018.03.007 · Cited by 25
MB-CNN: Memristive Binary Convolutional Neural Networks for Embedded Mobile Devices
Chowdhury, A. P., Kulkarni, P., & Bojnordi, M. N. (2018) · Journal of Low Power Electronics and Applications · doi:10.3390/jlpea8040038 · Cited by 25
Computation offloading for fast CNN inference in edge computing
Yang, Q., Luo, X., Li, P., Miyazaki, T., & Wang, X. (2019) · ACM Conference on Research in Adaptive and Convergent Systems (RACS) · doi:10.1145/3338840.3355669 · Cited by 25
“[18] introduce an analytical design policy for any CNN by using the roofline model.”
MethodSelf-citation
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.