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 801–900 of 1,002
TinyCNN: A Tiny Modular CNN Accelerator for Embedded FPGAPreprint
Jahanshahi, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.06777 · Cited by 21
“There are also efforts in the academic community on FPGA-based CNN accelerators [4,5] as well as tools for generating them automatically [6, 7].”
BackgroundFu, Y., Zhang, Y., Zhang, Y., Cox, D., & Lin, Y. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.06575 · Cited by 21
“…and (4) the network layer allocation, i.e., how to assign each layer to be processed by different chunks within the chunk-wise pipelined architecture, with all being critical accelerator parameters as pointed out by SOTA accelerator works (Chen et al., 2017; Zhang et al., 2015; Yang et al., 2016).”
BackgroundMethodology for CNN Implementation in FPGA-Based Embedded Systems
Zacchigna, F. G. (2022) · IEEE Embedded Systems Letters · doi:10.1109/les.2022.3187382 · Cited by 20
“In [2], [5], [8], and [9], the approach is to optimize the performance of CNNs by using design space exploration (DSE) and automatic code generation tools and in [8], a software API is also automatically generated, which provides an interface (IF) to configure, control, and monitor the design.”
Optimization of FPGA-based CNN accelerators using metaheuristics
Sait, S. M., El-Maleh, A. H., Altakrouri, M., & Shawahna, A. (2022) · The Journal of Supercomputing · doi:10.1007/s11227-022-04787-8 · Cited by 20
“In [16], the authors adopted the polyhedral-based data dependence analysis [27] to optimize the computations and memory access operations in CONV layers.”
BackgroundMethodInfluentialHardGBM: A Framework for Accurate and Hardware-Efficient Gradient Boosting Machines
Wang, H., Wu, Z., Wang, X., Bian, L., & Jin, H. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3218509 · Cited by 20
“CNN architecture design is adopted from [53], [54], [55].”
MethodHW-ADAM: FPGA-Based Accelerator for Adaptive Moment Estimation
Zhang, W., Niu, L., Zhang, D., Wang, G., Farrukh, F. U. D., & Zhang, C. (2023) · Electronics · doi:10.3390/electronics12020263 · Cited by 20
Self-citationReview on FPGA-Based Accelerators in Deep learning
Wu, Y. (2023) · IEEE Information Technology, Networking, Electronic and Automation Control Conference (ITNEC) · doi:10.1109/itnec56291.2023.10082175 · Cited by 20
“Researchers often consider the following parallelism: inter-output parallelism, inter-kernel parallelism, and intra-kernel Parallelism [1].”
BackgroundMethodInfluentialLi, J., Liu, C., Wang, L., Liu, Y., Li, R., Lu, X., Lü, J., & Shen, J. (2024) · Ecological Informatics · doi:10.1016/j.ecoinf.2024.102704 · Cited by 21
Curbing the roofline: a scalable and flexible architecture for CNNs on FPGA
Meloni, P., Deriu, G., Conti, F., Loi, I., Raffo, L., & Benini, L. (2016) · ACM International Conference on Computing Frontiers (CF) · doi:10.1145/2903150.2911715 · Cited by 21
“Table 4: Comparison of our work and [20] on two reference convolution layers”
BackgroundMethodResultInfluentialConvolutional neural network acceleration with hardware/software co-design
Chen, A. T.-Y., Biglari-Abhari, M., Wang, K. I., Bouzerdoum, A., & Tivive, F. H. C. (2017) · Applied Intelligence · doi:10.1007/s10489-017-1007-z · Cited by 20
“However, recent research [1, 18] has shown that as long as CNNs are trained with floating-point precision, the execution itself can be done using integer arithmetic without significantly harming accuracy.”
BackgroundA power-efficient and high performance FPGA accelerator for convolutional neural networks
Gong, L., Wang, C., Li, X., Chen, H., & Zhou, X. (2017) · International Conference on Hardware/Software Codesign and System Synthesis · doi:10.1145/3125502.3125534 · Cited by 20
“2015[5] ICCAD”
Exploiting Sparsity to Accelerate Fully Connected Layers of CNN-Based Applications on Mobile SoCs
Xie, X., Du, D., Li, Q., Liang, Y. E., Tang, W. T., Ong, Z., Lu, M., Huynh, H. P., & Goh, R. S. M. (2017) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3122788 · Cited by 20
“Zhang et al. accelerated deep convolutional networks on FPGAs and proposed a roofline-model for design space exploration (Zhang et al. 2015).”
BackgroundExact and Heuristic Allocation of Multi-kernel Applications to Multi-FPGA Platforms
Shan, J., Casu, M. R., Cortadella, J., Lavagno, L., & Lazarescu, M. T. (2019) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3316781.3317821 · Cited by 20
“As discussed in [10], we use loop tiling to reuse both the input feature maps and the weights.”
MethodThe challenge of multi-operand adders in CNNs on FPGAs: how not to solve it!
Abdelouahab, K., Pelcat, M., & Berry, F. (2018) · International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation · doi:10.1145/3229631.3235024 · Cited by 20
“Multiple state-of-the-art publications demonstrate the resiliency of CNNs towards compact bit-width arithmetic[2, 15] and even binarization[4, 12], which hints that CNNsmay support others types of approximate computing techniques such as approximate adders.”
BackgroundInfluentialEmbedded Intelligence in the Internet-of-Things
Dick, R. P., Shang, L., Wolf, M., & Yang, S. (2019) · IEEE Design and Test · doi:10.1109/mdat.2019.2957352 · Cited by 20
“FPGAs have been used to accelerate machine learning applications, including deep learning [98], [99].”
MethodOn the Resilience of Deep Learning for Reduced-voltage FPGAs
Givaki, K., Salami, B., Hojabr, R., Tayaranian, S. M. R., Khonsari, A., Rahmati, D., Gorgin, S., Cristal, A., & Ünsal, O. (2020) · Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP) · doi:10.1109/pdp50117.2020.00023 · Cited by 20
Chang, J.-W., Ahn, S., Kang, K.-W., & Kang, S. (2020) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac47756.2020.9045214 · Cited by 20
“We can decide the optimal tiling factors using the cross-layer optimization [21, 22].”
MethodFPGA-based neural network accelerators for millimeter-wave radio-over-fiber systems
Lee, J., He, J., & Wang, K. (2020) · Optics Express · doi:10.1364/oe.391050 · Cited by 20
“ations, the dominant computation units required in the CNN and BCNN hardware accelerators are different. In FPGA-based CNN hardware accelerators, 32-bit floating point numbers in the IEEE-754 standard [24] are used for computations, and hence, as shown in Figure 4(a), the DSP IP blocks are used for their hardware imple…”
MethodAhmad, H., Arif, T., Hanif, M. A., Hafiz, R., & Shafique, M. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3012865 · Cited by 20
“Zhang et al. [9] proposed a roofline-based analytical model that performs DSE to search for a design point with the best performance and lowest resource requirement, specifically for FPGAs.”
BackgroundMethodAn FPGA-Based Low-Latency Accelerator for Randomly Wired Neural Networks
Kuramochi, R., & Nakahara, H. (2020) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl50879.2020.00056 · Cited by 20
“Because of the limited number of on-chip memories on an FPGA and the different sizes of feature maps between the stages of RWNN, we use the tiling technique to process the CNN, as performed in [14] and [15].”
MethodDynamically Growing Neural Network Architecture for Lifelong Deep Learning on the Edge
Piyasena, D., Thathsara, M., Kanagarajah, S., Lam, S.-K., & Wu, M. (2020) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl50879.2020.00051 · Cited by 21
An Energy-Efficient Implementation of Group Pruned CNNs on FPGA
Pang, W., Wu, C., & Lu, S. (2020) · IEEE Access · doi:10.1109/access.2020.3041464 · Cited by 20
“However, general-purpose CNN accelerators [5]–[11] are still not efficient enough to process the compressed models.”
BackgroundDynamic Temperature Management of Near-Sensor Processing for Energy-Efficient High-Fidelity Imaging
Kodukula, V., Katrawala, S., Jones, B., Wu, C.-J., & LiKamWa, R. (2021) · Sensors · doi:10.3390/s21030926 · Cited by 20
“The state-of-the-art FPGA implementations [26,28,29] typically”
NLP-Fast: A Fast, Scalable, and Flexible System to Accelerate Large-Scale Heterogeneous NLP Models
Kim, J., Hur, S., Lee, E., Lee, S., & Kim, J. (2021) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1109/pact52795.2021.00013 · Cited by 20
“Some studies provide frameworks for automatic design space exploration on FPGA [69]–[75], and others try to maximize resource utilization [76], [76], [77].”
Optimizing Off-Chip Memory Access for Deep Neural Network Accelerator
Zheng, Y., Yang, H., Shu, Y., Jia, Y., & Huang, Z. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3150030 · Cited by 20
“Setting the size of ofmap buffer between 16KB to 512KB, we measured the DRAM access volume when using ofmap reuse oriented (ORO) scheduling scheme [8], weight reuse oriented (WRO) scheduling scheme [9] and our proposed adaptive scheduling scheme.”
BackgroundMethodHigh-Throughput and Power-Efficient Convolutional Neural Network Using One-Pass Processing Elements
Sivasankari, B., Shunmugathammal, M., Appathurai, A., & Kavitha, M. (2022) · Journal of Circuits Systems and Computers · doi:10.1142/s0218126622502267 · Cited by 20
A Hardware-Friendly Low-Bit Power-of-Two Quantization Method for CNNs and Its FPGA Implementation
Sui, X., Lv, Q., Bai, Y., Zhu, B., Zhi, L., Yang, Y., & Tan, Z. (2022) · Sensors · doi:10.3390/s22176618 · Cited by 20
“It can be observed in Table 11 that the traditional convolutional computation module using on-chip DSPs for multiplication operations consumes a large number of on-chip DSPs regardless of the bit width of weights.”
BackgroundMethodInfluentialHigh density, low energy, magnetic tunnel junction based block RAMs for memory-rich FPGAs
Tatsumura, K., Yazdanshenas, S., & Betz, V. (2016) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/fpt.2016.7929181 · Cited by 20
“In [3] for example, the authors detail a flow that optimizes convolutional neural network implementations on an FPGA by balancing computation and the required external memory bandwidth; generally the most efficient implementations must increase the computation required in order to cache enough data to stay below the e…”
BackgroundBroadening the exploration of the accelerator design space in embedded scalable platforms
Piccolboni, L., Mantovani, P., Guglielmo, G. D., & Carloni, L. P. (2017) · IEEE High Performance Extreme Computing Conference · doi:10.1109/hpec.2017.8091091 · Cited by 20
“[36] analyzed Convolutional Neural Networks (CNN) applications.”
CoNNA – Compressed CNN Hardware Accelerator
STRUHARIK, R. J. R., Vukobratović, B. Z., Erdeljan, A., & Rakanovic, D. (2018) · Euromicro Conference on Digital System Design (DSD) · doi:10.1109/dsd.2018.00070 · Cited by 20
“However, majority of these solutions are concerned with the acceleration of only one particular layer type from the CNN, typically the convolutional layer, because of its high computational demand, [8-12].”
BackgroundRV-CNN: Flexible and Efficient Instruction Set for CNNs Based on RISC-V Processors
Lou, W., Wang, C., Gong, L., & Zhou, X. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-29611-7_1 · Cited by 20
FPNet: Customized Convolutional Neural Network for FPGA Platforms
Yang, Y., Wang, C., Gong, L., & Zhou, X. (2019) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/icfpt47387.2019.00077 · Cited by 20
“The actual floatingpoint performance can not be higher than the minimum of Computational Roof or the CTC Ratio × BW [19].”
BackgroundCARLA: A Convolution Accelerator With a Reconfigurable and Low-Energy Architecture
Ahmadi, M., Vakili, S., & Langlois, J. M. P. (2021) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3066967 · Cited by 20
FLightNNs: Lightweight Quantized Deep Neural Networks for Fast and Accurate Inference
Ding, R., Liu, Z., Chin, T.-W., Marculescu, D., & Blanton, R. D. (. (2019) · Proceedings of the 56th Annual Design Automation Conference 2019 · doi:10.1145/3316781.3317828 · Cited by 19
An FPGA Based Energy Efficient DS-SLAM Accelerator for Mobile Robots in Dynamic Environment
Wu, Y., Luo, L., Yin, S., Yu, M., Qiao, F., Huang, H., Shi, X., Wei, Q., & Liu, X.-J. (2021) · Applied Sciences · doi:10.3390/app11041828 · Cited by 19
“For the design of FPGA, traditional development is mainly based on the RTL (Register Transfer Level) [12,13], which has a long development cycle and complicated debugging.”
MethodJi, M., Al-Ars, Z., Hofstee, H. P., Chang, Y., & Zhang, B. (2023) · Electronics · doi:10.3390/electronics12194085 · Cited by 19
Data Motion Acceleration: Chaining Cross-Domain Multi Accelerators
Wang, S., Xu, H., Mamandipoor, A., Mahapatra, R., Ahn, B. H., Ghodrati, S., Kailas, K. K., Alian, M., & Esmaeilzadeh, H. (2024) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca57654.2024.00083 · Cited by 19
Edge Computing Accelerator for Real-Time Defect Detection of Photovoltaic Panel on Lightweight FPGAs
Ye, J., Liu, Y., Chen, H., Wang, C., Zhou, Y., Yang, L., & Zhang, W. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3563001 · Cited by 19
“To identify the optimal BF for RDA, the roofline model [27] is leveraged, which provides insights into the tradeoffs between computational capacity and memory bandwidth.”
An Investigation Into the Efficacy of Deep Learning Tools for Big Data Analysis in Health Care
Priyadarshini, R., Barik, R. K., Panigrahi, C. R., Dubey, H., & Mishra, B. K. (2018) · International Journal of Grid and High Performance Computing · doi:10.4018/ijghpc.2018070101 · Cited by 19
A CNN Accelerator on FPGA with a Flexible Structure
Shan, D., Cong, G., & Lu, W. D. (2020) · International Conference on Computational Intelligence and Applications (ICCIA) · doi:10.1109/iccia49625.2020.00047 · Cited by 19
“Reference [14] implemented CNN using High Level Synthesis.”
MethodAn Efficient FPGA Accelerator Optimized for High Throughput Sparse CNN Inference
Wen, J., Ma, Y., & Wang, Z. (2020) · IEEE Asia Pacific Conference on Circuits and Systems (APCCAS) · doi:10.1109/apccas50809.2020.9301696 · Cited by 19
Korol, G., Jordan, M. G., Rutzig, M. B., & Beck, A. C. S. (2021) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3476990 · Cited by 19
“• Unlike works on statically generated CNN accelerators for FPGA [4, 13, 14, 28, 48] and works on FPGA-based Edge systems [26, 27, 43], it dynamically adapts the inference processing according to current quality and workload requirements.”
BackgroundMethodInfluentialBoosting Mobile CNN Inference through Semantic Memory
Li, Y., Zhang, C., Han, S., Zhang, L. L., Yin, B., Liu, Y., & Xu, M. (2021) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3474085.3475399 · Cited by 19
“Besides the aforementioned methods, extensive efforts have been made to optimize the CNN inference so they can be affordable on mobile/wearable devices, such as weight quantization [8, 10, 23], pruning [16, 30, 32, 36], hardware-based acceleration [50, 58, 59], model compression for mobile devices [28, 34, 53, 55, 60]…”
Self-citationReal-time task scheduling for FPGA-based multicore systems with communication delay
Xu, J., Li, K., & Chen, Y. (2022) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2022.104468 · Cited by 19
FPGA Accelerator for Homomorphic Encrypted Sparse Convolutional Neural Network Inference
Yang, Y., Kuppannagari, S. R., Kannan, R., & Prasanna, V. K. (2022) · IEEE International Symposium on Field-Programmable Custom Computing Machines (FCCM) · doi:10.1109/fccm53951.2022.9786115 · Cited by 19
SmartVM: A Smart Contract Virtual Machine for Fast On-Chain DNN Computations
Li, T., Fang, Y., Lu, Y., Yang, J., Jian, Z., Wan, Z., & Li, Y. (2022) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2022.3177405 · Cited by 19
FastStamp: Accelerating Neural Steganography and Digital Watermarking of Images on FPGAs
Hussain, S., Sheybani, N., Neekhara, P., Zhang, X., Duarte, J., & Koushanfar, F. (2022) · International Conference on Computer Aided Design · doi:10.1145/3508352.3549357 · Cited by 19
“To the best of our knowledge, we propose the first framework to accelerate deep neural networks for steganography and watermarking of images using FPGAs.”
BackgroundInfluentialDeep Learning on High Performance FPGA Switching Boards: Flow-in-Cloud
Musha, K., Kudoh, T., & Amano, H. (2018) · Lecture notes in computer science · doi:10.1007/978-3-319-78890-6_4 · Cited by 19
Swallow: A Versatile Accelerator for Sparse Neural Networks
Liu, B., Chen, X., Han, Y., & Xu, H. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.2978836 · Cited by 19
“D accelerator architectures [22], [23] that share activations across different PE rows, the differences of Swallow are in twofolds: 1) it exploits the weight and activation sparsity in SNN accelerations to reduce the memory traffic and ineffectual computation and 2) it enables data reuse of both…”
MethodReal-time FPGA-based Anomaly Detection for Radio Frequency Signals
Moss, D. J. M., Boland, D., Pourbeik, P., & Leong, P. H. W. (2018) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2018.8350890 · Cited by 19
“al in 2015, achieved 62 GFLOPS in single precision floating-point [4], using a roofline model to balance computational resources and memory bandwidth.”
MethodThe Implementation of CNN-Based Object Detector on ARM Embedded Platforms
Zhang, Y., Bi, S., Dong, M., & Liu, Y. (2018) · IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · doi:10.1109/dasc/picom/datacom/cyberscitec.2018.00074 · Cited by 19
Lin, X., Yin, S., Tu, F., Liu, L., Li, X., & Wei, S. (2018) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3195970.3196067 · Cited by 18
“However, in basic CONV tiling methods [10], tiling is performed within a layer, thus the inter-layer data always have to take circuitous routes, leave from computing array, pass through on-chip buffers and off-chip memory, and come back to computing array.”
BackgroundMethodInfluentialKowsalya, T. (2019) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2019.102906 · Cited by 18
Wang, G. F., Gu, C., Li, J., Wang, J., Chen, X., & Zhang, H. (2023) · Drones · doi:10.3390/drones7060380 · Cited by 18
“Typically, several accelerators for CNN have been proposed based on FPGA owing to its high performance and reconfigurability [104].”
MethodDeploying deep learning networks based advanced techniques for image processing on FPGA platform
Ghodhbani, R., Saidani, T. S., & Zayeni, H. (2023) · Neural Computing and Applications · doi:10.1007/s00521-023-08718-3 · Cited by 18
“[12], propose a set of optimizations that target the data storage, computation, and communication requirements of CNNs to maximize the resource utilization and minimize the power consumption of FPGA-based accelerators.”
BackgroundMobile Transformer Accelerator Exploiting Various Line Sparsity and Tile-Based Dynamic Quantization
Kwon, E., Yoon, J., & Kang, S. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3347291 · Cited by 19
“The SQ-GEMM performs matrix multiplication with the obtained dense matrices A’ [ ( M - α ) × N ] and B’ [ N × ( O - β )] based on tiling [33].”
MethodReconfigurable Stream Network Architecture
Wang, C. C., Zhang, X., Cong, J., & Hoe, J. C. (2025) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3695053.3731088 · Cited by 18
Self-citationSCALENet: A SCalable Low power AccELerator for Real-time Embedded Deep Neural Networks
Shea, C., Page, A., & Mohsenin, T. (2018) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3194554.3194601 · Cited by 18
“SCALENet delivers higher energy efficiency for a given network complexity than the previous best accelerators [5] and [16] while targeting a much lower power utilization.”
BackgroundMethodInfluentialDesign Methodology for Embedded Approximate Artificial Neural Networks
Balaji, A., Ullah, S., Das, A. K., & Kumar, A. S. (2019) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3299874.3319490 · Cited by 18
“Considering the flexibility, high-parallelism and reconfigurability of FPGAs, recent works have also considered FPGA-based ANN accelerators [6, 11, 13, 16, 17].”
MethodLin, C., & Lai, B.-C. C. (2018) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/aspdac.2018.8297290 · Cited by 18
“Specific accelerators [4–11] were proposed to speedup the intensive computation of CONV layers.”
MethodModel Parallelism Optimization for CNN FPGA Accelerator
Wang, J., Tong, W., & Zhi, X. (2023) · Algorithms · doi:10.3390/a16020110 · Cited by 18
An Energy-Efficient Reconfigurable Autoencoder Implementation on FPGA
Isik, M., Oldland, M., & Zhou, L. (2024) · Lecture notes in networks and systems · doi:10.1007/978-3-031-47721-8_14 · Cited by 18
A Multithreaded CGRA for Convolutional Neural Network Processing
Ando, K., Takamaeda-Yamazaki, S., Ikebe, M., Asai, T., & Motomura, M. (2017) · Circuits and Systems · doi:10.4236/cs.2017.86010 · Cited by 18
“analysed the data reusability and the arithmetic intensity of input- and output-parallel architectures built on an FPGA, and provided a method to optimize it when a target application is given [2].”
MethodNVM-Based FPGA Block RAM With Adaptive SLC-MLC Conversion
Ju, L., Sui, X., Li, S., Zhao, M., Xue, C. J., Hu, J., & Jia, Z. (2018) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2018.2857261 · Cited by 18
Scalable FPGA Accelerator for Deep Convolutional Neural Networks with Stochastic Streaming
Alawad, M., & Lin, M. (2018) · IEEE Transactions on Multi-Scale Computing Systems · doi:10.1109/tmscs.2018.2886266 · Cited by 18
“improve both, such as [21], these studies all follow the conventional way to realize CNNs and suffer from the high computational complexity, memory bandwidth, off-chip memory access, and inter-layers connections complexity challenges.”
InfluentialAn Efficient Accelerator Design Methodology For Deformable Convolutional Networks
Ahn, S., Chang, J.-W., & Kang, S. (2020) · Proceedings - International Conference on Image Processing · doi:10.1109/icip40778.2020.9191028 · Cited by 18
“Architecture Design for Preventing Irregular Memory Accesses Existing CNN architectures [21], [22] are inappropriate to accelerate DCL due to the random sampling process.”
MethodAn Energy Efficient EdgeAI Autoencoder Accelerator for Reinforcement Learning
Manjunath, N. K., Shiri, A., Hosseini, M., Prakash, B., Waytowich, N. R., & Mohsenin, T. (2021) · IEEE Open Journal of Circuits and Systems · doi:10.1109/ojcas.2020.3043737 · Cited by 18
“Using the rooftop model these three methods were tested in [35] to determine which provides the best throughput in FPGA.”
MethodDeepFense: Online Accelerated Defense Against Adversarial Deep LearningPreprint
Rouhani, B. D., Samragh, M., Javaheripi, M., Javidi, T., & Koushanfar, F. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1709.02538 · Cited by 18
High-density Integrated Photonic Tensor Processing Unit with a Matrix Multiply CompilerPreprint
Ma, X., Peserico, N., Khaled, A., Guo, Z., Nouri, B. M., Dalir, H., Shastri, B. J., & Sorger, V. J. (2022) · Research Square · doi:10.21203/rs.3.rs-1833027/v1 · Cited by 18
Song, L., Guo, L., Basalama, S., Chi, Y., Lucas, R. F., & Cong, J. (2023) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3543622.3573182 · Cited by 17
“tion processing on the fly? Many previous FPGA accelerators support a fixed-size problem such as deep learning accelerators [10, 41, 49], stencil computation [4, 5], graph convolutional network acceleration [48], and other applications [19].”
BackgroundSelf-citationAccelerating Convolutional Neural Network by Exploiting Sparsity on GPUs
Xu, W., Sun, Y., Fan, S., Yu, H., & Fu, X. (2023) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3600092 · Cited by 17
Meng, Y., Yang, C., Xiang, S., Wang, J., Mei, K., & Geng, L. (2023) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2023.3298509 · Cited by 17
“In [56], the throughput of the AlexNet conv1 (11 × 11 convolution) layer is just 32.8% of AlexNet conv2 (5 × 5 convolution) and 34.9% of AlexNet conv3 while VX485T is applied.”
MethodInfluentialKeep in Balance: Runtime-reconfigurable Intermittent Deep Inference
Yen, C.-H., Mendis, H. R., Kuo, T., & Hsiu, P.-C. (2023) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3607918 · Cited by 17
“As formulated in prior studies [19, 28, 35, 51], Nrd and Nwd depend on the DNN layer’s characteristics (e.”
Yang, J., Zheng, H., & Louri, A. (2023) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2023.3340953 · Cited by 17
Using Fermat number transform to accelerate convolutional neural network
Xu, W., You, X., & Zhang, C. (2017) · IEEE International Conference on ASIC (ASICON) · doi:10.1109/asicon.2017.8252655 · Cited by 17
“89 × resource efficiency (GOP/s/DSP) compared to [12] and [13].”
BackgroundSelf-citationLightening the Load with Highly Accurate Storage- and Energy-Efficient LightNNs
Ding, R., Liu, Z., Blanton, R. D., & Marculescu, D. (2018) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3270689 · Cited by 17
“[44], the loop col can be fully pipelined for achieving better performance.”
BackgroundInfluentialNeuralHMC: an efficient HMC-based accelerator for deep neural networks
Min, C., Mao, J., Li, H. H., & Chen, Y. (2019) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3287624.3287642 · Cited by 17
“It has been proven that the execution of a DNN is composed of many multiplications and additions, which can be accelerated us-ing dedicated accelerators [17].”
BackgroundSmall Memory Footprint Neural Network Accelerators
Seto, K., Nejatollahi, H., An, J., Kang, S., & Dutt, N. D. (2019) · International Symposium on Quality Electronic Design (ISQED) · doi:10.1109/isqed.2019.8697641 · Cited by 17
“[2] proposed a technique”
Symmetric $k$-Means for Deep Neural Network Compression and Hardware Acceleration on FPGAs
Jain, A., Goel, P., Aggarwal, S., Fell, A., & Anand, S. (2020) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2020.2968810 · Cited by 18
Alhichri, H. S., Bazi, Y., & Alajlan, N. (2020) · Arabian Journal for Science and Engineering · doi:10.1007/s13369-020-04799-7 · Cited by 17
“[63] analyzed the throughput and required memory bandwidth for various CNNs using optimization techniques, such as loop tiling and transformation.”
MethodFPGA Architecture Exploration for DNN Acceleration
Roorda, E., Rasoulinezhad, S. A., Leong, P. H. W., & Wilton, S. J. E. (2022) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3503465 · Cited by 17
“Early tools [46, 49] identify optimal dataflows and un-rolling factors, which users would then implement manually.”
MethodOffline Training-Based Mitigation of IR Drop for ReRAM-Based Deep Neural Network Accelerators
Lee, S., Fouda, M. E., Lee, J., Eltawil, A. M., & Kurdahi, F. J. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3177002 · Cited by 17
“In addition, we assume that convolution layers are mapped to RCAs by parallelizing the input channels and output channels [26].”
MethodSAMBA: S parsity A ware In- M emory Computing B ased Machine Learning A ccelerator
Kim, D. E., Ankit, A., Wang, C., & Roy, K. (2023) · IEEE Transactions on Computers · doi:10.1109/tc.2023.3257513 · Cited by 17
“Sze et al. [14] classifies ML accelerators based on their data handling characteristics: weight stationary architectures [15], [16], [17], [18], [19], output stationary architectures [20], [21], [22], architectures with no local reuse [23], [24], [25], and row stationary architectures [26].”
BackgroundA high-speed reusable quantized hardware accelerator design for CNN on constrained edge device
Yanamala, R. M. R., & Muralidhar, P. (2023) · Design Automation for Embedded Systems · doi:10.1007/s10617-023-09274-8 · Cited by 17
“It achieved 3.42GOP/s by consuming more DSPs. [45, 46] implemented the LeNet model using more resources (24 times more DSPs than ours) for getting more GOP/s and the power consumption is more here.”
MethodAn FPGA Realization of a Deep Convolutional Neural Network Using a Threshold Neuron Pruning
Fujii, T., Sato, S., Nakahara, H., & Motomura, M. (2017) · Lecture notes in computer science · doi:10.1007/978-3-319-56258-2_23 · Cited by 17
SOLAR: Services-oriented Deep Learning Architectures
Wang, C., Gong, L., Wang, A., Li, X., Hung, P. C. K., & Zhou, X. (2017) · IEEE Transactions on Services Computing · doi:10.1109/tsc.2017.2777478 · Cited by 17
“ral Networks (DNNs) [2] and Convolution Neural Networks (CNNs) [3], which have excellent capability in solving picture recognition, voice recognition, and other complex machine learning tasks.”
BackgroundA Non-Stop Double Buffering Mechanism for Dataflow Architecture
Tan, X., Shen, X.-W., Ye, X., Wang, D., Fan, D., Zhang, L., Li, W., Zhang, Z., & Tang, Z. (2018) · Journal of Computer Science and Technology · doi:10.1007/s11390-017-1747-6 · Cited by 17
Qiu, K., Chen, W., Xu, Y., Xia, L., Wang, Y., & Shao, Z. (2018) · Design, Automation, and Test in Europe · doi:10.23919/date.2018.8342168 · Cited by 17
“Compared to the manner of executing MAC operations in GPU, ASIC and FPGA [3], [4], the RRAM crossbar-based CNN algorithms [1] demonstrate two highlights.”
MethodFPGA Implementation of a Functional Neuro-Fuzzy Network for Nonlinear System Control
Jhang, J., Tang, K.-H., Huang, C.-K., Lin, C., & Young, K.-Y. (2018) · Electronics · doi:10.3390/electronics7080145 · Cited by 17
“Furthermore, the realization that a hybrid of neural networks and fuzzy systems presents an even more powerful form of computational intelligence [12] provides additional motivation to complete hardware implementation.”
BackgroundMethodInfluentialAccelerating a Secure Programmable Edge Network System for Smart Classroom
Nsunza, W. W., Tetteh, A.-Q. R., & Hei, X. (2018) · IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI) · doi:10.1109/smartworld.2018.00240 · Cited by 17
“[20] using a “Xilinx Virtex 7 485T”.”
Enabling Overclocking Through Algorithm-Level Error Detection
Marty, T., Yuki, T., & Derrien, S. (2018) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/fpt.2018.00034 · Cited by 17
“CNN models used in state-of-the-art applications are computation intensive and often need to be accelerated on GPUs or FPGAs to achieve high performance and/or obtain better energy efficiency [11], [12].”
Efficient Hardware Realization of Convolutional Neural Networks Using Intra-Kernel Regular Pruning
Yang, M., Faraj, M., Hussein, A., & Gaudet, V. (2018) · IEEE International Symposium on Multiple-Valued Logic (ISMVL) · doi:10.1109/ismvl.2018.00039 · Cited by 17
CNN-based Object Detection on Low Precision Hardware: Racing Car Case Study
Rita, N. D., Aimar, A., & Delbrück, T. (2019) · IEEE Intelligent Vehicles Symposium · doi:10.1109/ivs.2019.8814001 · Cited by 17
“Dedicated hardware accelerators gathered a lot of attention in recent years thanks to their high energy efficiency, able to outpace traditional GPU-based systems and many architectures have been proposed based for both ASIC and FPGA implementations [4]–[8].”
Optimization of Convolutional Neural Networks on Resource Constrained Devices
Sateesan, A., Sinha, S., & Smitha, K. G. (2019) · IEEE Computer Society Annual Symposium on VLSI · doi:10.1109/isvlsi.2019.00013 · Cited by 17
“b have the best FPS/DSP value compared to [3], [4], [22], [23].”
BackgroundBiS-KM: Enabling Any-Precision K-Means on FPGAs
He, Z., Wang, Z., & Alonso, G. (2020) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3373087.3375316 · Cited by 16
“Hardware acceleration of deep neural networks [27, 31, 47, 48, 50] and machine learning algorithms [21, 22, 29] has been a common topic for many years.”
BackgroundAccelerating End-to-End Deep Learning Workflow with Codesign of Data Preprocessing and Scheduling
Yang, C. H., Li, D., Guo, Z., Jiang, B., Geng, J., Lin, J., Fan, X., Yu, X., Bai, W., Qu, L., et al. (2020) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2020.3047966 · Cited by 17
“…DL studies are about the explorations of FPGAs on accelerating the specific computational workloads for fast inference of NNs, such as CNN [59], [60], LSTM [61], [62], etc. Differently, DLbooster improves the end-to-end DL workflows by offloading some key data preprocessing workloads to FPGAs.”
BackgroundShivapakash, S., Jain, H., Hellwich, O., & Gerfers, F. (2021) · IEEE Open Journal of Circuits and Systems · doi:10.1109/ojcas.2020.3047225 · Cited by 17
“5 × less power when compared to [21] and 4 × lesser when compared to [32] for AlexNet inference.”
BackgroundReliability of Neural Networks Based on Spintronic Neurons
Raimondo, E., Giordano, A., Grimaldi, A., Puliafito, V., Carpentieri, M., Zeng, Z., Tomasello, R., & Finocchio, G. (2021) · IEEE Magnetics Letters · doi:10.1109/lmag.2021.3100317 · Cited by 17
Yu, M., Huang, S., & Chen, D. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-91608-4_52 · Cited by 17
Lin, X., Yin, S., Tu, F., Liu, L., Li, X., & Wei, S. (2018) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac.2018.8465777 · Cited by 17
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.