The DOI matches its Crossref record; 144 of the 157 references it cites were found in a registry or an index.
Efficient Processing of Deep Neural Networks: A Tutorial and Survey
Abstract
Deep neural networks (DNNs) are currently widely used for many artificial intelligence (AI) applications including computer vision, speech recognition, and robotics. While DNNs deliver state-of-the-art accuracy on many AI tasks, it comes at the cost of high computational complexity. Accordingly, techniques that enable efficient processing of DNNs to improve energy efficiency and throughput without sacrificing application accuracy or increasing hardware cost are critical to the wide deployment of DNNs in AI systems. This article aims to provide a comprehensive tutorial and survey about the recent advances toward the goal of enabling efficient processing of DNNs. Specifically, it will provide an overview of DNNs, discuss various hardware platforms and architectures that support DNNs, and highlight key trends in reducing the computation cost of DNNs either solely via hardware design changes or via joint hardware design and DNN algorithm changes. It will also summarize various development resources that enable researchers and practitioners to quickly get started in this field, and highlight important benchmarking metrics and design considerations that should be used for evaluating the rapidly growing number of DNN hardware designs, optionally including algorithmic codesigns, being proposed in academia and industry. The reader will take away the following concepts from this article: understand the key design considerations for DNNs; be able to evaluate different DNN hardware implementations with benchmarks and comparison metrics; understand the tradeoffs between various hardware architectures and platforms; be able to evaluate the utility of various DNN design techniques for efficient processing; and understand recent implementation trends and opportunities.
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
“Classical arithmetic operations, such as MAC, constitute the bulk of the computational requirements for models used in artificial intelligence [13].”
From Brain-inspired computing systems: a systematic literature review (2024) · cites this paper as [13]
In the citing paper
“To clarify, consider the multiply-and-accumulate (MAC) operation in digital processors, where the operation energy consumption is significantly dependent on the bit width of the input operands [13, 235] or equivalently to the signal-to-noise ratio (SNR) in analog computing [236].”
From Brain-inspired computing systems: a systematic literature review (2024) · cites this paper as [13]
In the citing paper
“Altogether, processing one 224 × 224 input image needs a VGG-16 model of 15.5 G multiply-and-accumulates and 138 M weights [40].”
From Face Recognition Based on Deep Learning and FPGA for Ethnicity Identification (2022) · cites this paper as [40]
In the citing paper
“The energy consumption of executing a DNN model is highly dictated by the computation architecture and methods (e.g., ultra-low power compute-in-memory AI accelerator) at the edge computation nodes [249], the architecture of DNN models [250], and the wireless transmission for data exchange during the model inference procedure.”
From Edge artificial intelligence for 6g: vision, enabling technologies, and applications (2022) · cites this paper as [250]
In the citing paper
“but there is no review discussing all three categories which in future will be integrated seamlessly to design effective and efficient DNNs in the new edge intelligence era.”
From Bringing ai to edge: from deep learning’s perspective (2022) · cites this paper as [25]
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 15×
“In 2015, the ImageNet winning entry, ResNet [15], exceeded human-level accuracy with a top-5 error rate4 below 5%.”
InfluentialVery Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 10×
“Used in VGG-16. decompose 5x5 filter 5x1 filter 1x5 filter Apply sequentially (b) Constructing a 5×5 support from 1×5 and 5×1 filter.”
InfluentialImageNet 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 10×
“this, local response normalization (LRN) [3] was used, which”
BackgroundMethodInfluentialHochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×
“In contrast, recurrent neural networks (RNNs), of which Long Short-Term Memory networks (LSTMs) [38] are a popular variant, have internal memory to allow long-term dependencies to affect the output.”
LeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
Gradient-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 4×
“Table 2 Summary of Popular DNNs [3], [15], [48], [50], [51]”
InfluentialGoing deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380 · cited 8×
“†ACCURACY IS MEASURED BASED ON TOP-5 ERROR ON IMAGENET [14].”
InfluentialFully convolutional networks for semantic segmentation
Long, J., Shelhamer, E., & Darrell, T. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298965 · Cited by 38,834
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.81 · Cited by 32,684
Research works printed but not linked: 34
- 10.13140/RG.2.2.18893.74727 · DOI · Paper page
- Li, Stanford CS Class CS231n: Convolutional Neural Networks for Visual Recognition
- 2016, Complete Visual Networking Index (VNI) Forecast
- Woodhouse, 2016, Big, Big, Big Data: Higher and Higher Resolution Video Surveillance
- Maas, Rectifier nonlinearities improve neural network acoustic models, 1
- Caffe LeNet MNIST
- Caffe Model Zoo
- Matconvnet Pretrained Models
- TensorFlow-Slim Image Classification Library
- Deep Learning Frameworks
- Yann, The MNIST DATABASE of Handwritten Digits
- Krizhevsky, The CIFAR-10 Dataset
- Krizhevsky, 2010, Convolutional deep belief networks on cifar-10, Unpublished Manuscript, 40, 1
- Pascal VOC Data Sets
- Microsoft Common Objects in Context (COCO) Dataset
- Google Open Images
- YouTube-8M
- AudioSet
- 2017, Intel Architecture Instruction Set Extensions Programming Reference
- Condon, Facebook unveils Big Basin, new server geared for deep learning
- Intel Math Kernel Library
- 10.1145/1816038.1815993 · DOI
- 10.1145/2872887.2750389 · DOI
- 10.1145/2644865.2541967 · DOI · Paper page
- Standard, High bandwidth memory (HBM) DRAM
- 10.1145/3093336.3037702 · DOI
- Wilson, International technology roadmap for semiconductors (ITRS)
- 10.1109/icvd.2004.1260946 · doi:10.1109/icvd.2004.1260946
- Lee, A 2.5 GHz 7.7 TOPS/W switched-capacitor matrix multiplier with co-designed local memory in 40 nm, 418
- Lee, LogNet: Energy-efficient neural networks using logrithmic computations, 5900
- DNN Energy Estimation
- 10.1145/3140659.3080254 · DOI · Paper page
- Iandola, SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <1 MB model size
- Benchmarking DNN Processors
Cited by
Items 901–1,000 of 1,011
Agrawal, A., Kosta, A. K., Kodge, S., Kim, D. E., & Roy, K. (2020) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2020.3014250 · Cited by 25
Ozen, E., & Orailoğlu, A. (2020) · doi:10.1145/3400302.3415680 · Cited by 25
“The ability to handle fault tolerance at low cost and for large numbers of faults introduces the possibility of underpinning innovative DNN accelerator optimizations in performance and power, an area of intense research focus [32].”
BackgroundResultReconfigurable Binary Neural Network Accelerator with Adaptive Parallelism Scheme
Cho, J., Jung, Y., Lee, S., & Jung, Y. (2021) · Electronics · doi:10.3390/electronics10030230 · Cited by 27
“These functions are usually conventional nonlinear functions, such as the rectified linear unit (ReLU), sigmoid, and hyperbolic tangent [6].”
BackgroundDoes AI have a hardware problem?
(2018) · Nature Electronics · doi:10.1038/s41928-018-0068-2 · Cited by 25
Wang, Z., Kara, K., Zhang, H., Alonso, G., Mutlu, O., & Zhang, C. (2019) · Proceedings of the VLDB Endowment · doi:10.14778/3317315.3317322 · Cited by 30
Searching for Winograd-aware Quantized NetworksPreprint
Fernández-Marqués, J., Whatmough, P. N., Mundy, A., & Mattina, M. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.10711 · Cited by 45
“Lower-precision networks result in smaller model sizes, faster inference, lower energy consumption and smaller chip area (Sze et al., 2017; Li et al., 2019).”
BackgroundFrenkel, C., Bol, D., & Indiveri, G. (2021) · Zurich Open Repository and Archive (University of Zurich) · doi:10.5167/uzh-217499 · Cited by 24
A visual introduction to Gaussian Belief PropagationPreprint
Ortiz, J., Evans, T., & Davison, A. J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.02308 · Cited by 47
“This design has the key aim of increasing performance while reducing power usage by minimizing the ”bits x millimetres” of data movement during computation [52].”
BackgroundZhang, H., Wang, H., Xu, Z., Liu, Z., & Khoo, B. C. (2023) · Journal of Wind Engineering and Industrial Aerodynamics · doi:10.1016/j.jweia.2023.105534 · Cited by 23
EdgeCI: Distributed Workload Assignment and Model Partitioning for CNN Inference on Edge Clusters
Chen, Y., Luo, T., Fang, W. W., & Xiong, N. N. (2024) · ACM Transactions on Internet Technology · doi:10.1145/3656041 · Cited by 27
“…workload partition (excepting for margin) will be expanded along all the dimensions to contain the data required for convolution operation whose kernel size is greater than 1 in fused block f m , and the outward offset of each workload partition is determined by the size of convolution kernel [10].”
BackgroundMethodInfluentialSelf-citationAdvanced deep learning approaches in metasurface modeling and design: A review
Dong, Y., An, S., Jiang, H., Zheng, B., Tang, H. X., Huang, Y., Zhao, H., & Zhang, H. (2025) · Progress in Quantum Electronics · doi:10.1016/j.pquantelec.2025.100554 · Cited by 28
Benchmark of Ferroelectric Transistor-Based Hybrid Precision Synapse for Neural Network Accelerator
Luo, Y., Wang, P., Peng, X., Sun, X., & Yu, S. (2019) · IEEE Journal on Exploratory Solid-State Computational Devices and Circuits · doi:10.1109/jxcdc.2019.2925061 · Cited by 25
Kolenov, D., & Pereira, S. F. (2020) · Optics Express · doi:10.1364/oe.395233 · Cited by 26
“Throughout the IC manufacturing process, large amounts of data need to be mined in a fully automated mode [38].”
BackgroundJeon, J., Park, S., Choi, M., Kim, J., Kwon, Y., & Cho, S. (2020) · Electronics · doi:10.3390/electronics9091359 · Cited by 24
“To deal with the larger workloads, data centers have implemented distributed neural network training techniques [1].”
BackgroundKim, M., & Seo, J.-S. (2020) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2020.3029235 · Cited by 25
“For energy-efficient DCNN accelerators, it is crucial to design a computing scheme that can suppress redundant computations, exploit data sparsity statistics, and optimize system-level memory hierarchy that can reduce the amount of off-chip memory accesses [7].”
Shomron, G., Banner, R., Shkolnik, M., & Weiser, U. C. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58607-2_14 · Cited by 24
Flash Memory Array for Efficient Implementation of Deep Neural Networks
Han, R., Xiang, Y., Huang, P. C., Shan, Y., Liu, X., & Kang, J. (2020) · Advanced Intelligent Systems · doi:10.1002/aisy.202000161 · Cited by 23
Xia, Z., Chen, J., Huang, Q., Luo, J., & Hu, J. (2020) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2020.3039346 · Cited by 27
BBNet: A Novel Convolutional Neural Network Structure in Edge-Cloud Collaborative Inference
Zhou, H., Zhang, W., Wang, C., Ma, X., & Yu, H. (2021) · Sensors · doi:10.3390/s21134494 · Cited by 26
“Its impact on accuracy will not exceed 1% [27].”
BackgroundArtificial Neural Networks to Forecast Failures in Water Supply Pipes
Velasco, A. R., Ramos-Salgado, C., Muñuzuri, J., & Cortés, P. (2021) · Sustainability · doi:10.3390/su13158226 · Cited by 26
Towards Edge Computing Using Early-Exit Convolutional Neural Networks
Pacheco, R. G., Bochie, K., Gilbert, M. S., Couto, R. S., & Campista, M. E. M. (2021) · Information · doi:10.3390/info12100431 · Cited by 30
“As a consequence, more computational power is needed, representing a challenge for resource-constrained devices [4].”
BackgroundScaling Deep-Learning Inference with Chiplet-based Architecture and Photonic Interconnects
Li, Y., Louri, A., & Karanth, A. (2021) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac18074.2021.9586311 · Cited by 23
“, row-stationary dataflow [3]), SPRINT dataflow may change accordingly.”
BackgroundEfficientTDNN: Efficient Architecture Search for Speaker Recognition
Wang, R., Wei, Z., Duan, H., Ji, S., Long, Y., & Hong, Z. (2022) · IEEE/ACM Transactions on Audio Speech and Language Processing · doi:10.1109/taslp.2022.3182856 · Cited by 28
“The inference time complexity of a subnet O(a) can be summarized as Theorem 1, regardless of the number of nonzero weights and nonzero MACs that reflect the theoretical minimum requirements of storage and computation [35].”
BackgroundEfficient Fault-Criticality Analysis for AI Accelerators using a Neural Twin
Chaudhuri, A., Chen, C.-Y., Talukdar, J., Madala, S., Dubey, A. K., & Chakrabarty, K. (2021) · IEEE International Test Conference (ITC) · doi:10.1109/itc50571.2021.00015 · Cited by 24
Radiomics, deep learning and early diagnosis in oncology
Wei, P. (2021) · Emerging Topics in Life Sciences · doi:10.1042/etls20210218 · Cited by 27
A Practical Design-Space Analysis of Compute-in-Memory With SRAM
Spetalnick, S. D., & Raychowdhury, A. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3138057 · Cited by 30
“The energy cost of modern MAC workloads consists of digital logic and memory access components, the latter having become more important in recent years [1], [24].”
BackgroundVariable-Precision Approximate Floating-Point Multiplier for Efficient Deep Learning Computation
Zhang, H., & Ko, S. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3161005 · Cited by 23
Spiking Neural Network-Based Radar Gesture Recognition System Using Raw ADC Data
Arsalan, M., Santra, A., & Іссаков, В. (2022) · IEEE Sensors Letters · doi:10.1109/lsens.2022.3173589 · Cited by 23
Weak self-supervised learning for seizure forecasting: a feasibility study
Yang, Y., Truong, N. D., Eshraghian, J. K., Nikpour, A., & Kavehei, O. (2022) · Royal Society Open Science · doi:10.1098/rsos.220374 · Cited by 26
Improvement of Min-Entropy Evaluation Based on Pruning and Quantized Deep Neural Network
Li, H., Zhang, J., Li, Z., Liu, J., & Wang, Y. (2023) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2023.3240859 · Cited by 28
“The main idea of network pruning is to remove a certain percentage of relatively unimportant parameters without affecting the network performance [31].”
BackgroundAn Empirical Approach to Enhance Performance for Scalable CORDIC-Based Deep Neural Networks
Raut, G., Karkun, S., & Vishvakarma, S. K. (2023) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3596220 · Cited by 32
Haq, S. A. U., Gizzini, A. K., Shrey, S., Darak, S. J., Saurabh, S., & Chafii, M. (2023) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2023.3274555 · Cited by 27
“These functions include Sigmoid, tanh, ReLU, leaky ReLU [49].”
BackgroundNonlinear optical feature generator for machine learning
Yildirim, M., Oğuz, İ., Kaufmann, F., Escalé, M. R., Grange, R., Psaltis, D., & Moser, C. (2023) · APL Photonics · doi:10.1063/5.0158611 · Cited by 39
DeAR: Accelerating Distributed Deep Learning with Fine-Grained All-Reduce Pipelining
Zhang, L., Shi, S., Chu, X., Wang, W., Li, B., & Liu, C. (2023) · IEEE International Conference on Distributed Computing Systems (ICDCS) · doi:10.1109/icdcs57875.2023.00054 · Cited by 44
“It is worth pointing out that feed-forward computations account for around one third of the total computation time in each iteration [18], which can be properly exploited to further accelerating the training speed.”
BackgroundMethodInfluentialVulnerability assessment of heat waves within a risk framework using artificial intelligence
Côté, J.-N., Germain, M., Levac, E., & Lavigne, É. (2023) · The Science of The Total Environment · doi:10.1016/j.scitotenv.2023.169355 · Cited by 26
Reusing Deep Learning Models: Challenges and Directions in Software Engineering
Davis, J. C., Jajal, P., Jiang, W.-X., Schorlemmer, T. R., Synovic, N. M., & Thiruvathukal, G. (2023) · John Vincent Atanasoff Symposium · doi:10.1109/jva60410.2023.00015 · Cited by 37
“Once the weights are finalized by minimizing the loss function, the loss function itself is no longer required. d) Optimizations: Based on the device being used for training, the model training time and efficiency will vary drastically [36], [37].”
MethodSwapNet: Efficient Swapping for DNN Inference on Edge AI Devices Beyond the Memory Budget
Wang, K., Cao, J., Zhou, Z., & Li, Z. (2024) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2024.3355764 · Cited by 27
“We illustrate the memory budget for DNN inference by developing a self-driving application on standard edge AI devices below.”
MethodEfficient Spiking Neural Networks with Biologically Similar Lithium-Ion Memristor Neurons
Ke, S., Pan, Y., Jin, Y., Meng, J., Xiao, Y., Chen, S., Zhang, Z., Li, R., Tong, F., Jiang, B., et al. (2024) · ACS Applied Materials & Interfaces · doi:10.1021/acsami.3c19261 · Cited by 24
Sharma, A., Haq, S. A. U., & Darak, S. J. (2024) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2024.3371780 · Cited by 26
“CNN accelerators aim to reduce DDR access by leveraging the data reuse properties of CONV layers [43].”
MethodMogaka, O. M., Zewail, R., Inoue, K., & Sayed, M. S. (2024) · Journal of Real-Time Image Processing · doi:10.1007/s11554-024-01430-y · Cited by 23
“In addition, the hardware implementation of APoT leverages bitwise shift operations, renowned for their energy efficiency and suitability for on-chip memory and in-memory computing [41].”
Xu, Y., Shidqi, K., van Schaik, G.-J., Bilgic, R., Dobrita, A., Wang, S., Meijer, R., Nembhani, P. V., Arjmand, C., Martinello, P., et al. (2024) · Frontiers in Neuroscience · doi:10.3389/fnins.2024.1335422 · Cited by 30
Segmentation technique for medical image processing: A survey
Merjulah, R., & Chandra, J. (2017) · International Conference on Inventive Computing and Informatics (ICICI) · doi:10.1109/icici.2017.8365301 · Cited by 24
“[2] Vivienne Sze, Yu-Hsin Chen, Tien-Ju Yang, and Joel Emer, “Efficient Processing of Deep Neural Networks: A Tutorial and Survey”, in computer vision and pattern recognition, eprint arXiv:1703.”
BackgroundReconfigurable Convolutional Kernels for Neural Networks on FPGAs
Hardieck, M., Kumm, M., Möller, K., & Zipf, P. (2019) · ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA) · doi:10.1145/3289602.3293905 · Cited by 31
“typically large amounts of data are processed using the same coefficients (often thousands of convolutions, see Section 2) by using the weight-stationary scheme [28], this short reconfiguration time is often negligible.”
Automated Circuit Approximation Method Driven by Data Distribution
Vašíček, Z., Mrázek, V., & Sekanina, L. (2019) · Design, Automation, and Test in Europe · doi:10.23919/date.2019.8714977 · Cited by 24
“With the rapid development of artificial intelligence methods based on deep CNNs, a lot of attention has been focused on efficient hardware implementations of neural networks [10].”
MethodRethinking Pruning for Accelerating Deep Inference At the Edge
Gao, D., He, X., Zhou, Z., Tong, Y., Xu, K., & Thiele, L. (2020) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3394486.3403058 · Cited by 30
“One effective technique to accelerate deep neural networks is network pruning [26, 32].”
BackgroundMethodSECDA: Efficient Hardware/Software Co-Design of FPGA-based DNN Accelerators for Edge Inference
Haris, J., Gibson, P., Cano, J., Agostini, N. B., & Kaeli, D. R. (2021) · International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD) · doi:10.1109/sbac-pad53543.2021.00015 · Cited by 49
A role for analogue memory in AI hardware
Burr, G. W. (2018) · Nature Machine Intelligence · doi:10.1038/s42256-018-0007-y · Cited by 22
Peng, Y., Zhong, W., Peng, Z., Tu, Y., Xu, Y., Li, Z., Liang, J., Huang, J., Liu, X., & Fu, Y. (2024) · Agronomy · doi:10.3390/agronomy14061248 · Cited by 22
Reducing Energy Bloat in Large Model Training
Chung, J.-W., Gu, Y., Jang, I., Meng, L., Bansal, N., & Chowdhury, M. (2024) · ACM SIGOPS Symposium on Operating Systems Principles (SOSP) · doi:10.1145/3694715.3695970 · Cited by 66
“While energy optimization is well-studied in the hardware community [7, 14, 75, 84], the power bottleneck of recent datacenters [9, 11, 12, 53, 65] shows that efficiency gains from hardware advancement alone are not sufficient to sustain the growing demand for AI compute.”
Edge Deployment of Deep Networks for Visual Detection: A Review
Zhai, H., Du, J., Ai, Y., & Hu, T. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3502539 · Cited by 35
“Model pruning can significantly reduce the model size and computational load without compromising or only a slight decrease in accuracy, thus improving the efficiency and usability of the model [95], [96], [97].”
Li, T. J., Cao, Y., Ye, Q., & Zhang, Y. (2025) · Journal of Manufacturing Processes · doi:10.1016/j.jmapro.2025.02.053 · Cited by 24
An, Y., Kim, D., Park, Y. R., Eo, J. S., Kim, M., Kim, D., Kim, H. B., Lee, B., & Wang, G. (2025) · Nano Energy · doi:10.1016/j.nanoen.2025.110999 · Cited by 23
Adversarial denoising of EEG signals: a comparative analysis of standard GAN and WGAN-GP approaches
Tibermacine, I. E., Russo, S., Citeroni, F., Mancini, G., Rabehi, A., Alharbi, A. H., El‐kenawy, E. M., & Napoli, C. (2025) · Frontiers in Human Neuroscience · doi:10.3389/fnhum.2025.1583342 · Cited by 38
Feature Allocation for Semantic Communication With Space-Time Importance Awareness
Zhou, K., Zhang, G., Cai, Y., Hu, Q., Yu, G., & Swindlehurst, A. L. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3569320 · Cited by 29
Helaly, M. A., Rady, S., Mabrouk, M., Aref, M. M., Villarroya, S., Cotos, J. M., & Mera, D. (2025) · Cluster Computing · doi:10.1007/s10586-025-05221-3 · Cited by 22
“Next was the investigation of the impact of employing a simple Convolutional Neural Network (CNN) architecture on a water quality dataset to predict pH levels [2]— mathematical foundations are details in [19].”
A systematic review of AI-driven business models for advancing Sustainable Development Goals
Khan, M. A., Rahman, A., Mahmud, F. U., Bishnu, K. K., Ahmed, M., Mridha, M. F., & Aung, Z. (2025) · Array · doi:10.1016/j.array.2025.100539 · Cited by 34
Neural networks: Efficient implementations and applications
Zhang, C., & Xu, W. (2017) · IEEE International Conference on ASIC (ASICON) · doi:10.1109/asicon.2017.8252654 · Cited by 22
Balance gate controlled deep neural network
Liu, A., & Laili, Y. (2018) · Neurocomputing · doi:10.1016/j.neucom.2018.08.075 · Cited by 25
Energy-Efficient Architecture for CNNs Inference on Heterogeneous FPGA
Spagnolo, F., Perri, S., Frustaci, F., & Corsonello, P. (2019) · Journal of Low Power Electronics and Applications · doi:10.3390/jlpea10010001 · Cited by 23
“float32x4_t vexp[125], L[125]; float32x4_t vexp2[125], L2[125]; float32x4_t sumexp; float totsum=0; for (j=0; j<125; j=j+1){ for (i=0; i<4; i=i+1){ vexp[j][i]=exp((simdsum8[j][i])>>nf); vexp2[j][i]=exp((simdsum8[j][i+4]) >>nf); } sumexp=vaddq_f32(vexp[j],vexp2[j]); totsum=totsum+sumexp[0]+sumexp[1]+sumexp[2]+sumexp[3]…”
BackgroundChewBaccaNN: A Flexible 223 TOPS/W BNN Accelerator
Andri, R., Karunaratne, G., Cavigelli, L., & Benini, L. (2021) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas51556.2021.9401214 · Cited by 27
Modeling of Pruning Techniques for Simplifying Deep Neural Networks
Pasandi, M. M., Hajabdollahi, M., Karimi, N., & Samavi, S. (2020) · Iranian Conference on Machine Vision and Image Processing · doi:10.1109/mvip49855.2020.9116891 · Cited by 42
“CNNs are used in a lot of applications such as image processing, pattern recognition, speech recognition, etc. [1].”
BackgroundLi, T., Ma, Y., & Endoh, T. (2020) · IEEE Access · doi:10.1109/access.2020.3013934 · Cited by 22
Artificial Neural Network Model for Design Optimization of 2-stage Op-amp
Harsha, M. V., & Harish, B. P. (2020) · International Symposium on VLSI Design and Test (VDAT) · doi:10.1109/vdat50263.2020.9190315 · Cited by 22
“…the presence of high degree of nonlinearity and complex relationships between multiple performance metrics and design parameters, Multi-layer Perceptron (MLP) with sufficiently large number of hidden layers and neurons are chosen to effectively map input-output relationships [5], [14], [15], [20].”
BackgroundMethodHigh Performance and Portable Convolution Operators for Multicore Processors
Juan, P. S., Castelló, A., Dolz, M. F., Alonso, P., & Quintana–Ort́ı, E. S. (2020) · International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD) · doi:10.1109/sbac-pad49847.2020.00023 · Cited by 23
“…to a myriad of unexplored applications, for example, in quantum computing, solid state lighting, nanoelectronics and nanomechanics, high throughput screening of new materials, computer vision in microscopy, radiography and tomography, and astrophysics simulation; see [4]– [6] among many others.”
BackgroundMethodBagchi, S., Mohd, M. N. H., Debnath, S. K., Marwan Nafea, M., Suriani, N. S., & Nizam, Y. (2020) · IEEE Student Conference on Research and Development (SCOReD) · doi:10.1109/scored50371.2020.9250939 · Cited by 23
CPU-Accelerator Co-Scheduling for CNN Acceleration at the Edge
Kim, Y., Kong, J., & Munir, A. (2020) · IEEE Access · doi:10.1109/access.2020.3039278 · Cited by 23
“Recently, as CNNs have gained a huge attention due to their wide range of AI applications, it calls for hardware accelerators [21] [22] that execute CNNs with better performance and energy efficiency.”
BackgroundDeep Neural Network-Based Semantic Segmentation of Microvascular Decompression Images
Bai, R., Jiang, S., Sun, H., Yang, Y., & Li, G. (2021) · Sensors · doi:10.3390/s21041167 · Cited by 23
Genetic Programming-Based Discriminative Feature Learning for Low-Quality Image Classification
Bi, Y., Xue, B., & Zhang, M. (2021) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2021.3049778 · Cited by 24
“The deep features are from the pre-trained models of VGG, Xception, InceptionV3, ResNet, and DenseNet, respectively [57].”
BackgroundMethodInfluentialAntolini, A., Scarselli, E. F., Gnudi, A., Carissimi, M., Pasotti, M., Romele, P., & Canegallo, R. A. (2021) · Materials · doi:10.3390/ma14071624 · Cited by 22
On Machine-Learning Morphological Image Operators
Hirata, N. S. T., & Papakostas, G. A. (2021) · Mathematics · doi:10.3390/math9161854 · Cited by 25
Efficient Approximate Multiplier Based on a New 1-Gate Approximate Compressor
Ejtahed, S. A. H., & Timarchi, S. (2022) · Circuits Systems and Signal Processing · doi:10.1007/s00034-021-01902-7 · Cited by 22
“For example, 3.9 × 10 9 multiply-and-accumulate operations have been used in ResNet-50 [26].”
MethodInfluentialMultimodal transistors as ReLU activation functions in physical neural network classifiers
Pesch, I. S., Bestelink, E., de Sagazan, O., Mehonić, A., & Sporea, R. A. (2022) · UCL Discovery (University College London) · Cited by 22
Chen, C.-Y., Lai, A. I.-C., Wu, P.-Y., & Wu, R. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3147644 · Cited by 25
Resources and Power Efficient FPGA Accelerators for Real-Time Image Classification
Kyriakos, A., Papatheofanous, E. A., Bezaitis, C., & Reisis, D. I. (2022) · Journal of Imaging · doi:10.3390/jimaging8040114 · Cited by 24
“proposed designs are to avoid idle computing and memory [21] resources, use only the on-chip (FPGA) memory and extensive pipeline, features that lead to improved resources utilization, reduced latency and power consumption [22,23].”
BackgroundSurgeNAS: A Comprehensive Surgery on Hardware-Aware Differentiable Neural Architecture Search
Luo, X., Liu, D., Kong, H., Huai, S., Chen, H., & Liu, W. (2022) · IEEE Transactions on Computers · doi:10.1109/tc.2022.3188175 · Cited by 23
“…[19] have achieved remarkable breakthroughs among various real-world scenarios, but, unfortunately, such breakthroughs come at the cost of extremely high computational complexity [34] since tens of millions of floating-point operations (FLOPs) and prohibitive memory accesses (MACs) are involved.”
BackgroundGQNA: Generic Quantized DNN Accelerator With Weight-Repetition-Aware Activation Aggregating
Yang, J., Tu, F., Li, Y., Wang, Y., Liu, L., Wei, S., & Yin, S. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3188899 · Cited by 22
“With the same number of quantization levels, NUQ-weights can approximate the normal distribution of floating-point weights better than UQ-weights, due to unequal quantization intervals [17], [20]–[22].”
BackgroundFedDQ: A communication-efficient federated learning approach for Internet of Vehicles
Mo, Z., Gao, Z., Zhao, C., & Lin, Y. (2022) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2022.102690 · Cited by 25
Distributed Inference in Resource-Constrained IoT for Real-Time Video Surveillance
Khan, M. A., Hamila, R., Erbad, A., & Gabbouj, M. (2022) · IEEE Systems Journal · doi:10.1109/jsyst.2022.3198711 · Cited by 29
“Usually, accessing dynamic random access memory (DRAM) requires 1 or 2 order of energy higher than accessing on-chip memory [18].”
BackgroundLiao, J., Widmer, L., Wang, X., Mauro, A. D., Nason, S. R., Chestek, C. A., Benini, L., & Jang, T. (2022) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas54282.2022.9869846 · Cited by 24
“While neural networks are powerful, they come at the cost of high computational complexity, implying high energy consumption in hardware implementation [10].”
BackgroundAssessment of Speckle-Pattern Quality using Deep-Learning-Based CNN
Kwon, T., Park, J., Jeong, H., & Park, K. (2022) · Experimental Mechanics · doi:10.1007/s11340-022-00906-x · Cited by 24
Fang, C., Ma, H., & Li, J. (2022) · Infrared Physics & Technology · doi:10.1016/j.infrared.2022.104483 · Cited by 24
Xie, Y., Hou, X., Guo, Y., Wang, X., & Zheng, J. (2023) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2023.3286072 · Cited by 23
“Given these real-world resource constraints, model efficiency becomes increasingly important for object detection [8], [9].”
BackgroundMao, R., Tang, L., Yuan, X., Ye, L., & Zhou, J. (2024) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca57654.2024.00023 · Cited by 33
“The existing designs mostly adopt the traditional output stationary (OS) dataflow to accelerate the spatiotemporal computations due to the irregularity and time-dependency of the spike [22].”
MethodLogarithmic number system for deep learning
Kouretas, I., & Paliouras, V. (2018) · International Conference on Modern Circuits and Systems Technologies · doi:10.1109/mocast.2018.8376572 · Cited by 23
“Hardware implementations are of great interest for DNN computation acceleration with low-complexity energy-efficient and highthroughput architectures [9].”
BackgroundImage quality enhancement of a CD-SEM image using conditional generative adversarial networks
Midoh, Y., & Nakamae, K. (2019) · Metrology, Inspection, and Process Control XXXVI · doi:10.1117/12.2515152 · Cited by 22
“In recent years, deep learning has become very popular in solving many high level vision problems, such as denoising, super-resolution, segmentation, and inpainting [3].”
BackgroundQuTiBench: Benchmarking Neural Networks on Heterogeneous Hardware
Blott, M., Halder, L., Leeser, M. E., & Doyle, L. (2019) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3358700 · Cited by 33
“A more detailed description can be found in Reference [75].”
BackgroundSpArNet: Sparse Asynchronous Neural Network execution for energy efficient inference
Khoei, M. A., Yousefzadeh, A., Pourtaherian, A., Moreira, O., & Tapson, J. C. (2020) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas48895.2020.9073827 · Cited by 31
“Nonetheless, the performance of CNNs is mainly described by accuracy metrics for the given task, and less attention is devoted to efficient and optimized implementation of them [10].”
BackgroundComparing Different Decodings for Posit Arithmetic
Murillo, R., Mallasén, D., Barrio, A. A. D., & Botella, G. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-09779-9_6 · Cited by 22
Weight-Sharing Neural Architecture Search: A Battle to Shrink the Optimization GapPreprint
Xie, L., Chen, X., Bi, K., Wei, L., Xu, Y., Chen, Z., Wang, L., Xiao, A., Chang, J., Zhang, X., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.01475 · Cited by 29
“Two of the most popular techniques are network pruning [409], [410], [411], [412] that eliminates less useful”
MethodWu, P.-Y., Chou, W.-C., Wu, X. W., Kamineni, V. N., Kuchimanchi, Y., Tell, L. A., Maunsell, F. P., & Lin, Z. (2024) · Toxicological Sciences · doi:10.1093/toxsci/kfae125 · Cited by 23
An Approximate Multiply-Accumulate Unit with Low Power and Reduced Area
Yang, T., Sato, T., & Ukezono, T. (2019) · IEEE Computer Society Annual Symposium on VLSI · doi:10.1109/isvlsi.2019.00076 · Cited by 24
“It performs large numbers of multiply-accumulate (MAC) operations and therefore consumes significant power [4].”
BackgroundSelf-citationAutomatic reduction of wireless capsule endoscopy reviewing time based on factorization analysis
Biniaz, A., Zoroofi, R. A., & Sohrabi, M. (2020) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2020.101897 · Cited by 22
Learned Resolution Scaling Powered Gaming-as-a-Service at Scale
Chen, H., Lu, M., Ma, Z., Zhang, X., Xu, Y., Shen, Q., & Zhang, W. (2020) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2020.2985538 · Cited by 21
“Giventhefastgrowthofneural processingtechniques,weexpectthattheopticalflowcalcula-tioncanbesoonacceleratedbydedicatedNPU[43].”
BackgroundMixed-Signal Vector-by-Matrix Multiplier Circuits Based on 3D-NAND Memories for Neurocomputing
Bavandpour, M., Sahay, S., Mahmoodi, M. R., & Strukov, D. B. (2020) · Design, Automation, and Test in Europe · doi:10.23919/date48585.2020.9116401 · Cited by 25
Dinelli, G., Meoni, G., Rapuano, E., Pacini, T., & Fanucci, L. (2020) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2020.3015294 · Cited by 22
“This Streamline architecture limits the accesses to the external memory since only the input image shall be read, and the final results shall be stored, drastically reducing latency and power consumption [4], [32].”
BackgroundHolistic Filter Pruning for Efficient Deep Neural Networks
Enderich, L., Timm, F., & Burgard, W. (2021) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv48630.2021.00264 · Cited by 23
“But, recent progress has been made by training powerful models with many parameters using large scale data sets [8, 32].”
BackgroundMethodA Mura Detection Model Based on Unsupervised Adversarial Learning
Song, S., Yang, K., Wang, A., Zhang, S., & Xia, M. (2021) · IEEE Access · doi:10.1109/access.2021.3069466 · Cited by 24
“This structure is called bottleneck [35], the purpose is to reduce the amount of network parameters and the consumption of computing resources.”
BackgroundTowards Fully 8-bit Integer Inference for the Transformer Model
Lin, Y., Li, Y., Liu, T., Xiao, T., Liu, T., & Zhu, J. (2020) · Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2020/520 · Cited by 30
“Beyond this, INT8 is 10 × more energy efficient [ Johnson, 2018 ] and saves much less chip area than the commonly used 32-bit floating point (FP32) in hardware design [ Sze et al. , 2017 ] .”
MethodEdge ai: leveraging the full potential of deep learning
Shuvo, M. M. H. (2022) · Studies in Computational Intelligence · doi:10.1007/978-3-031-14748-7_2 · Cited by 18
Yan, Z., Juan, D.-C., Hu, X. S., & Shi, Y. (2021) · Proceedings of the 26th Asia and South Pacific Design Automation Conference · doi:10.1145/3394885.3431635 · Cited by 18
“CiM architectures offer reduced data movement by in-situ weight data access [17].”
Background
Metadata
- Authors
- Sze, V., Chen, Y.-H., Yang, T.-J., & Emer, J. S.
- Affiliations
- Massachusetts Institute of Technology (US)
- Nvidia (United States) (US)
- Venue
- Proceedings of the IEEE
- Published
- 2017-12
- Volume
- 105
- Issue
- 12
- Pages
- 2295-2329
- ISSN
- 0018-9219 (ISSN-L)
- 1558-2256
- OpenAlex
- W2604319603
- Cited by
- 4,215
- References
- 157
- Citations by year
- 2026: 504 · 2025: 523 · 2024: 474 · 2023: 466 · 2022: 508 · 2021: 568 · 2020: 551 · 2019: 392 · 2018: 199 · 2017: 28
- FWCI
- 132.75
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Proceedings of the IEEE (publishedVersion)
- Funders
- DARPA YFA
- MIT CICS
- gift from Nvidia
- gift from Intel
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
- Keywords
- deep neural network, energy efficiency, throughput, hardware accelerators, computation cost reduction, benchmarking metrics
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.