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 701–800 of 1,011
In-Hardware Training Chip Based on CMOS Invertible Logic for Machine Learning
Onizawa, N., Smithson, S. C., Meyer, B. H., Gross, W. J., & Hanyu, T. (2019) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2019.2960383 · Cited by 36
“As summarized in [45], recent developments of efficient hardware on deep neural networks are limited to inference only, and do not include training-on-chip.”
BackgroundEfficient Winograd or Cook-Toom Convolution Kernel Implementation on Widely Used Mobile CPUs
Mullins, R., Maji, P., Mundy, A., Dasika, G., Beu, J., & Mattina, M. (2019) · Apollo (University of Cambridge) · doi:10.17863/cam.40388 · Cited by 32
Practical Attacks on Deep Neural Networks by Memory Trojaning
Hu, X., Zhao, Y. K., Deng, L., Liang, L., Zuo, P., Ye, J., Lin, Y., & Xie, Y. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.2995347 · Cited by 43
“Deep Neural Networks (DNNs) have achieved extraordinary accuracy for many tasks, such as computer vision, speech recognition, and machine translation [1].”
BackgroundLiu, S., Du, J., Nan, K., Zhou, Z., Liu, H., Wang, Z., & Lin, Y. (2020) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2020.2999956 · Cited by 41
“Compression is a commonly employed technique to trim down the complexity of DNNs, which can be performed by reducing the weight precision, or the number of operations, or both [6].”
BackgroundMethodPutra, R. V. W., Hanif, M. A., & Shafique, M. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2021.3060509 · Cited by 46
“The redundant accesses to DRAM-based off-chip memory hinder the accelerators from achieving further efficiency gains as DRAM energy is significantly higher compared to other operations [25]–[29], which is also shown in Fig.”
ResultReal-time Automatic Modulation Classification using RFSoC
Tridgell, S., Boland, D., Leong, P. H. W., Kastner, R., Khodamoradi, A., & Siddhartha (2020) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw50202.2020.00021 · Cited by 37
“[15] for a comprehensive overview of modern CNN implementation).”
BackgroundDeep learning-based question answering system for intelligent humanoid robot
Budiharto, W., Andreas, V., & Gunawan, A. A. S. (2020) · Journal Of Big Data · doi:10.1186/s40537-020-00341-6 · Cited by 36
“They predict the span by using the bilinear term to capture the similarity between paragraphs and questions.”
MethodThe viability of analog-based accelerators for neuromorphic computing: a survey
Musisi‐Nkambwe, M., Afshari, S., Barnaby, H., Kozicki, M. N., & Esqueda, I. S. (2021) · Neuromorphic Computing and Engineering · doi:10.1088/2634-4386/ac0242 · Cited by 36
“Various popular models exist (e.g. LeNet5 [14], AlexNet [15], VGG [16], GoogleNet [17], ResNet50 [18], Bert-99 [19]) and are compared and tabulated in [4].”
BackgroundMethodInfluentialAslam, A. R., & Altaf, M. A. B. (2021) · IEEE Transactions on Biomedical Circuits and Systems · doi:10.1109/tbcas.2021.3113613 · Cited by 34
ETA: An Efficient Training Accelerator for DNNs Based on Hardware-Algorithm Co-Optimization
Lu, J., Ni, C., & Wang, Z. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3145850 · Cited by 36
Intelligent energy storage management trade-off system applied to Deep Learning predictions
Cordeiro-Costas, M., Villanueva, D., Eguía, P., & Granada, E. (2023) · Journal of Energy Storage · doi:10.1016/j.est.2023.106784 · Cited by 39
Human worker activity recognition in a production floor environment through deep learning
Mastakouris, A., Andriosopoulou, G., Masouros, D., Benardos, P., Vosniakos, G.-C., & Soudris, D. (2023) · Journal of Manufacturing Systems · doi:10.1016/j.jmsy.2023.08.020 · Cited by 32
Deep neural network based environment sound classification and its implementation on hearing aid app
Fan, X., Sun, T., Chen, W., & Fan, Q. (2020) · Measurement · doi:10.1016/j.measurement.2020.107790 · Cited by 33
Muneeswaran, V., Nagaraj, P., & Ijaz, M. F. (2022) · Studies in computational intelligence · doi:10.1007/978-3-030-97929-4_8 · Cited by 35
Proteus: A High-Throughput Inference-Serving System with Accuracy Scaling
Ahmad, S., Guan, H., Friedman, B. D., Williams, T., Sitaraman, R., & Woo, T. Y. C. (2024) · International Conference on Architectural Support for Programming Languages and Operating Systems · doi:10.1145/3617232.3624849 · Cited by 63
“Figure 1a shows the accuracy of EfficientNet variants [40] and their throughput in queries per second (QPS) on three different types of devices.”
BackgroundImage-based intrusion detection system for GPS spoofing cyberattacks in unmanned aerial vehicles
Korium, M. S., Saber, M., Ahmed, A. M., Narayanan, A., & Nardelli, P. H. J. (2024) · Ad Hoc Networks · doi:10.1016/j.adhoc.2024.103597 · Cited by 37
A high-throughput reconfigurable processing array for neural networks
Wu, E., Zhang, X., Berman, D., & Cho, I. (2017) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.23919/fpl.2017.8056794 · Cited by 37
“Because each output channel is pre-assigned to a particular processing column, partial sums are accumulated only at the top of each column (considered to be output-stationary in [27]).”
BackgroundMethodInfluentialPerformance analysis of real-time DNN inference on Raspberry Pi
Velasco-Montero, D., Fernández‐Berni, J., Carmona‐Galán, R., & Rodriguez-Vazquez, A. (2018) · doi:10.1117/12.2309763 · Cited by 35
“The interest of both industry and academia in DL is spurring the development of a myriad of software environments and hardware platforms tailored for DNNs.(6) Likewise, new network models are reported almost on a daily basis.”
BackgroundLearning-Assisted Optimization for Energy-Efficient Scheduling in Deadline-Aware NOMA Systems
Lei, L., You, L., He, Q., Vu, T. X., Chatzinotas, S., Yuan, D., & Ottersten, B. (2019) · IEEE Transactions on Green Communications and Networking · doi:10.1109/tgcn.2019.2902838 · Cited by 37
“…in real-time operations, the complexity of feed-forward operations of the DNN is mainly from the matrix multiplication, e.g., weight matrix times input matrix, the activation function, and the learning function [25], which are of low complexity in general, and can be executed fast in practice [26].”
BackgroundMethodBuilding Robust Machine Learning Systems: Current Progress, Research Challenges, and Opportunities
Zhang, J. J., Liu, K., Khalid, F., Hanif, M. A., Rehman, S., Theocharides, T., Artusi, A., Shafique, M., & Garg, S. (2019) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3316781.3323472 · Cited by 32
“Apart from the aforementioned-objectives, the digital system design community is focusing on developing efficient hardware accelerators which can further boost the efficiency gains by designing application-specific hardware for DNN-based applications [25].”
BackgroundSpeeding-up convolutional neural networks: A survey
Lebedev, V., & Lempitsky, V. (2018) · Bulletin of the Polish Academy of Sciences Technical Sciences · doi:10.24425/bpas.2018.125927 · Cited by 38
Deep Learning for Cardiovascular Risk Stratification
Schlesinger, D. E., & Stultz, C. M. (2020) · Current Treatment Options in Cardiovascular Medicine · doi:10.1007/s11936-020-00814-0 · Cited by 34
Mauro, A. D., Conti, F., Schiavone, P. D., Rossi, D., & Benini, L. (2020) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2020.3012576 · Cited by 34
“trend by becoming more and more specialized to efficiently execute machine learning workloads [17], [28].”
BackgroundVita, A. D., Russo, A., Pau, D. P., Benedetto, L. D., Rubino, A., & Licciardo, G. D. (2020) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2020.3011984 · Cited by 33
“Since the energy cost of data read/write operations from off-chip memories can be up to 200× higher than on chip data transfer [19], an effort has been done to use only on-chip memories.”
BackgroundDeep Learning Based Pothole Detection and Reporting System
Ganesh Babu, R., Chellaswamy, C., Surya Bhupal Rao, M., Saravanan, M., Kanchana, E., & Shalini, J. (2020) · International Conference on Smart Structures and Systems (ICSSS) · doi:10.1109/icsss49621.2020.9202061 · Cited by 31
“A deep learning method (DLM) and its efficiency have studied by Sze et al. [14].”
MethodMachine Learning Techniques used for the Histopathological Image Analysis of Oral Cancer-A Review
Panigrahi, S., & Swarnkar, T. (2020) · The Open Bioinformatics Journal · doi:10.2174/1875036202013010106 · Cited by 39
Evolutionary Multi-Objective Model Compression for Deep Neural Networks
Wang, Z., Luo, T., Li, M., Zhou, J. T., Goh, R. S. M., & Zhen, L. (2021) · IEEE Computational Intelligence Magazine · doi:10.1109/mci.2021.3084393 · Cited by 38
“Network Pruning and Quantization For making the training easy, the networks are usually overparameterized [14].”
BackgroundImproving Semantic Segmentation of Urban Scenes for Self-Driving Cars with Synthetic Images
Ivanovs, M., Ozols, K., Dobrājs, A., & Kadiķis, R. (2022) · Sensors · doi:10.3390/s22062252 · Cited by 36
“However, despite their impressive performance on many perceptual tasks, DNNs also have shortcomings: their inner workings are not transparent, as they operate as black boxes [19,20], they require a lot of computational power for running inference [9,21], and training them takes a lot of time, computing resources, espe…”
BackgroundHybrid Accumulator Factored Systolic Array for Machine Learning Acceleration
Inayat, K., & Chung, J. (2022) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2022.3170233 · Cited by 31
DORIS: Personalized course recommendation system based on deep learning
Ma, Y., Ouyang, R., Xinzheng, L. O. N. G., Gao, Z., Lai, T., & Fan, C. (2023) · PLoS ONE · doi:10.1371/journal.pone.0284687 · Cited by 46
Chae, J. (2024) · IEEE Open Journal of the Solid-State Circuits Society · doi:10.1109/ojsscs.2024.3458900 · Cited by 34
“However, the energy consumption of data transfer in memory interfaces accounts for a significant portion of the total energy consumption in AI operations [18]; therefore, the need for improved energy efficiency of memory interfaces continues to grow.”
Performance enhancement of artificial intelligence: A survey
Krichen, M., & Abdalzaher, M. S. (2024) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2024.104034 · Cited by 41
Deep Learning Methods in Soft Robotics: Architectures and Applications
Čakurda, T., Trojanová, M., Pomin, P., & Hošovský, A. (2024) · Advanced Intelligent Systems · doi:10.1002/aisy.202400576 · Cited by 34
Intelligent Edge Computing and Machine Learning: A Survey of Optimization and Applications
Cajas, S. A., Samanta, J., Suárez‐Cetrulo, A. L., & Carbajo, R. S. (2025) · Future Internet · doi:10.3390/fi17090417 · Cited by 56
“To tackle footprint concerns, deep compression techniques through pruning unnecessary connections and intelligent weight distribution have gained significant attention [22,23].”
Vinayaraj, P., Wang, W., & Nakamura, R. (2019) · Remote Sensing · doi:10.3390/rs11242961 · Cited by 30
“Deep learning networks deliver state-of-the-art accuracy on many images and point cloud segmentation tasks, but it comes at the cost of high computational complexity [42].”
BackgroundMachine Learning based prediction of noncentrosymmetric crystal materials
Song, Y., Lindsay, J., Zhao, Y., Nasiri, A., Louis, S.-Y., Ling, J., Hu, M., & Hu, J. (2020) · Computational Materials Science · doi:10.1016/j.commatsci.2020.109792 · Cited by 31
Deep Photometric Stereo Networks for Determining Surface Normal and Reflectances
Santo, H., Samejima, M., Sugano, Y., Shi, B., & Matsushita, Y. (2020) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2020.3005219 · Cited by 30
“5 M (in the BRDF estimation block) ≈ 8 M, which is regarded a small model size compared to the modern CNN architectures such as AlexNet, VGG-16 (used in GeoNet as backbone), and ResNet-50 [49].”
MethodTo Filter Prune, or to Layer Prune, That Is the Question
Elkerdawy, S., Elhoushi, M., Singh, A. K., Zhang, H., & Ray, N. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-69535-4_45 · Cited by 33
CRIME: Input-Dependent Collaborative Inference for Recurrent Neural Networks
Pagliari, D. J., Chiaro, R., Macii, E., & Poncino, M. (2020) · IEEE Transactions on Computers · doi:10.1109/tc.2020.3021199 · Cited by 36
“The spread of DL has been partially fueled by the proliferation of IoT sensors and smart devices generating massive amounts of data [2], [3].”
BackgroundDuplo: Lifting Redundant Memory Accesses of Deep Neural Networks for GPU Tensor Cores
Kim, H., Ahn, S., Oh, Y., Kim, B., Ro, W. W., & Song, W. J. (2020) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro50266.2020.00065 · Cited by 40
“Consequently, sliding filters on multi-dimensional input data results in deeply nested loops in every convolutional layer [40].”
BackgroundLocal Search is a Remarkably Strong Baseline for Neural Architecture Search
Ottelander, T. D., Dushatskiy, A., Virgolin, M., & Bosman, P. A. N. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-72062-9_37 · Cited by 51
Hanson, E., Li, S., Li, H. H., & Chen, Y. (2022) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3470496.3527419 · Cited by 30
“As discussed Sze et al. [33], the required precision for partial sums is typically 26 − 32 bits for popular DNNs.”
BackgroundSelf-citationAscend: A Scalable and Energy-Efficient Deep Neural Network Accelerator With Photonic Interconnects
Li, Y., Wang, K., Zheng, H., Louri, A., & Karanth, A. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3169953 · Cited by 30
Modeling Wildfire Spread with an Irregular Graph Network
Jiang, W., Wang, F., Su, G., Li, X., Wang, G., Zheng, X., Wang, T., & Meng, Q. (2022) · Fire · doi:10.3390/fire5060185 · Cited by 30
Perspective: Uniform switching of artificial synapses for large-scale neuromorphic arrays
Tan, S. H., Lin, P., Yeon, H., Choi, S., Park, Y., & Kim, J. (2018) · APL Materials · doi:10.1063/1.5049137 · Cited by 33
Cryptocurrency Trading: A Comprehensive SurveyPreprint
Fang, F., Ventre, C., Basios, M., Kanthan, L., Li, L., Martínez‐Rego, D., & Wu, F. (2020) · RePEc: Research Papers in Economics · doi:10.48550/arxiv.2003.11352 · Cited by 30
“Deep learning algorithms are currently the basis for many modern artificial intelligence applications [220].”
BackgroundReal-Time Semantic Segmentation: A brief survey and comparative study in remote sensing
Broni-Bediako, C., Xia, J., & Yokoya, N. (2023) · IEEE Geoscience and Remote Sensing Magazine · doi:10.1109/mgrs.2023.3321258 · Cited by 33
Domain-Specific STT-MRAM-Based In-Memory Computing: A Survey
Yusuf, A., Adegbija, T., & Gajaria, D. (2024) · IEEE Access · doi:10.1109/access.2024.3365632 · Cited by 34
“Current approaches such as graphics processing units (GPUs) [2] or application-specific processors [3], [4] are unlikely to overcome the problem of data movement because they are designed to perform specific tasks or functions and are not intended to The associate editor coordinating the review of…”
BackgroundEsmaeilpour, A., Tabarestani, S. S., & Niazi, A. (2024) · Engineering Reports · doi:10.1002/eng2.12918 · Cited by 34
“19 Its capacity to comprehend and manipulate data has significantly enhanced.”
Ahmad, F., Alasskar, A., Samui, P., & Asteris, P. G. (2025) · Frontiers of Structural and Civil Engineering · doi:10.1007/s11709-025-1201-8 · Cited by 39
Kwon, D., Malaiya, R. K., Yoon, G., Ryu, J.-T., & Pi, S.-Y. (2019) · Applied Sciences · doi:10.3390/app9142941 · Cited by 36
“The reason for calling it the simplest DNN model is that the DNN can be implemented by solely fully-connected layers [26].”
BackgroundMethodInfluentialNetwork Intrusion Detection Using Neural Networks on FPGA SoCs
Ioannou, L., & Fahmy, S. A. (2019) · International Conference on Field Programmable Logic and Applications (FPL) · doi:10.1109/fpl.2019.00043 · Cited by 30
“Network pruning, data quantization, and reduced arithmetic precision can be exploited to trade off performance, power consumption, and detection accuracy [23, 24, 25].”
BackgroundDeep Neural Network Architecture: Application for Facial Expression Recognition
Villanueva, M. G., & Zavala, S. R. (2020) · IEEE Latin America Transactions · doi:10.1109/tla.2020.9099774 · Cited by 31
“de las capas genera en forma sucesiva un nivel de abstracción superior de los datos de entrada, denominado mapa de caracterı́sticas, en el cual se preserva información única y esencial de las imágenes [27].”
BackgroundBurel, S., Evans, A., & Anghel, L. (2022) · IEEE Transactions on Device and Materials Reliability · doi:10.1109/tdmr.2022.3159089 · Cited by 35
Efficient optical reservoir computing for parallel data processing
Bu, T., Zhang, H., Kumar, S., Jin, M., Kumar, P., & Huang, Y. (2022) · Optics Letters · doi:10.1364/ol.464288 · Cited by 33
SRAM-Based Computing-in-Memory Macro With Fully Parallel One-Step Multibit Computation
Choi, E. J., Choi, I., Jeon, C., Yun, G., Yi, D., Ha, S., Chang, I., & Je, M. (2022) · IEEE Solid-State Circuits Letters · doi:10.1109/lssc.2022.3206416 · Cited by 31
“…recently shown ways to perform DNN computations in the mixed-signal domain with higher energy efficiency than the conventional Von-Neumann digital computation architectures [1], but they are prone to suffer from the limitations in their classification accuracy resulting from the analog nonidealities.”
MethodYüzügüler, A. C., Sönmez, C., Drumond, M. P., Oh, Y., Falsafi, B., & Frossard, P. (2022) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3572917 · Cited by 39
“Our design space exploration is isopower because DNN inference accelerators are typically bound by power consumption because of their dense arithmetic formats [61].”
BackgroundMethodA Survey of FPGA-Based Vision Systems for Autonomous Cars
Castells‐Rufas, D., Ngo, V., Borrego-Carazo, J., Codina, M., Sanchez, C., Gil, D., & Carrabina, J. (2022) · IEEE Access · doi:10.1109/access.2022.3230282 · Cited by 36
“Many of these tasks are increasingly implemented with DeepLearning (DL) approaches [26], and more particularly, using deep Convolutional Neural Networks (CNNs) [27].”
BackgroundKoziel, S. M., Çalık, N., Mahoutı, P., & Belen, M. A. (2022) · IEEE Transactions on Antennas and Propagation · doi:10.1109/tap.2022.3216064 · Cited by 29
“Some of the recent alternatives include deep learning (DL) algorithms [58], especially in the form of deep neural networks (DNNs) [54], which seem to be more efficient in the handling of nonlinear outputs of high-frequency components over conventional approaches [59], [60].”
BackgroundMaji, S., Banerjee, U., Fuller, S. H., & Chandrakasan, A. P. (2022) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2022.3215670 · Cited by 38
“Therefore, it is not surprising that a large number of NN hardware accelerators are designed on ASIC platforms [4].”
BackgroundLi, B., Liu, T., Liao, J., Feng, C., Yao, L., & Zhang, J. (2023) · Measurement · doi:10.1016/j.measurement.2023.112554 · Cited by 32
Unconventional computing based on magnetic tunnel junction
Cai, B., He, Y., Xin, Y., Yuan, Z., Zhang, X., Zhu, Z., & Liang, G. (2023) · Applied Physics A · doi:10.1007/s00339-022-06365-4 · Cited by 41
“In DNN, neurons in one layer are fully connected by neurons in the neighboring layer and information would be delivered unidirectionally.”
BackgroundMethodInfluentialLHAR: Lightweight Human Activity Recognition on Knowledge Distillation
Deng, S., Chen, J., Teng, D., Yang, C., Chen, D., Jia, T., & Wang, H. (2023) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2023.3298932 · Cited by 38
A novel deep learning model to predict the soil nutrient levels (N, P, and K) in cabbage cultivation
Sajindra, H., Abekoon, T., Jayakody, A., & Rathnayake, U. (2023) · Smart Agricultural Technology · doi:10.1016/j.atech.2023.100395 · Cited by 38
Ada3D : Exploiting the Spatial Redundancy with Adaptive Inference for Efficient 3D Object Detection
Zhao, T., Ning, X., Hong, K., Qiu, Z., Lu, P., Zhao, Y., Zhang, L., Zhou, L., Dai, G., Yang, H., et al. (2023) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51070.2023.01625 · Cited by 32
Naeini, S. S., & Snaiki, R. (2024) · Coastal Engineering · doi:10.1016/j.coastaleng.2024.104503 · Cited by 48
“Although DNNs are quite similar to ANN models, they are characterized by a deep architecture consisting of several hidden layers (Sze et al., 2017).”
MethodA Survey on Hardware Accelerator Design of Deep Learning for Edge Devices
Samanta, A., Hatai, I., & Mal, A. K. (2024) · Wireless Personal Communications · doi:10.1007/s11277-024-11443-2 · Cited by 39
“Sze et al., [40] aims to give an outline of DNNs, including several resources, data representations, hardware platforms, joint computational algorithms, and optimization techniques for increasing throughput and energy efficiency without jeopardising accuracy or driving up hardware costs, which is…”
Xu, C., Zhang, Y., Isleem, H. F., Qiu, D., Zhang, Y., Alsaadawi, M. M., Tipu, R. K., El‐Demerdash, W. E., & Hamed, A. (2024) · Structural Concrete · doi:10.1002/suco.202400541 · Cited by 39
Jia, H., Li, X., Zhuang, Y., Wu, Y., Shi, S., Sun, Q., He, F., Liang, S., Wang, J., Draz, M. S., et al. (2024) · Antimicrobial Agents and Chemotherapy · doi:10.1128/aac.01446-24 · Cited by 37
Reyes, J. M. R., Ho, I. W., & Mak, M. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.125802 · Cited by 44
The Escalating AI’s Energy Demands and the Imperative Need for Sustainable Solutions
León, M. (2024) · WSEAS TRANSACTIONS on SYSTEMS archive · doi:10.37394/23202.2024.23.46 · Cited by 35
Huang, J., Chen, J., Mai, H., Wan, H., Chen, R., & He, T. (2025) · Applied Acoustics · doi:10.1016/j.apacoust.2025.110633 · Cited by 27
Towards responsible AI for education: Hybrid human-AI to confront the elephant in the room
Hooshyar, D., Šír, G., Yang, Y., Kikas, E., Hämäläinen, R., Kärkkäinen, T., Gašević, D., & Azevedo, R. (2025) · Computers and Education Artificial Intelligence · doi:10.1016/j.caeai.2025.100524 · Cited by 54
Orrù, G., Marcialis, G. L., & Roli, F. (2019) · Pattern Recognition · doi:10.1016/j.patcog.2019.107121 · Cited by 31
“Another possibility is trying to rely on deep networks-based approaches [5], but even in this case, we must be able to incrementally train the whole net as novel samples and novel users are available.”
BackgroundLow-Power HWAccelerator for AI Edge-Computing in Human Activity Recognition Systems
Vita, A. D., Pau, D. P., Parrella, C., Benedetto, L. D., Rubino, A., & Licciardo, G. D. (2020) · IEEE International Conference on Artificial Intelligence Circuits and Systems (AICAS) · doi:10.1109/aicas48895.2020.9073913 · Cited by 30
“Number and delay of memory accesses are major bottlenecks for the performance of neural networks, in particular when accesses have to go through an off-chip DRAM [19].”
BackgroundA NoC-based simulator for design and evaluation of deep neural networks
Chen, K.-C. J., Ebrahimi, M., Wang, T., Yang, Y.-C., & Liao, Y.-H. (2020) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2020.103145 · Cited by 28
“The notable benefits of artificial neural networks (ANNs) have led to the advancement in many real-world applications, such as speech recognition and image classification [1] .”
BackgroundBerestizshevsky, K., & Even, G. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-30484-3_26 · Cited by 38
Architecture of neural processing unit for deep neural networks
Lee, K. J. (2020) · Advances in computers · doi:10.1016/bs.adcom.2020.11.001 · Cited by 30
Efficient activation functions for embedded inference engines
Wuraola, A., Patel, N. D., & Nguang, S. K. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.02.030 · Cited by 28
Zhou, J., Wang, J., Yang, C., Li, X., & Xie, Y. C. (2021) · Sensors · doi:10.3390/s21217271 · Cited by 29
Multimodal transistors as ReLU activation functions in physical neural network classifiers
Pesch, I. S., Bestelink, E., de Sagazan, O., Mehonić, A., & Sporea, R. A. (2022) · Scientific Reports · doi:10.1038/s41598-021-04614-9 · Cited by 31
“Using MATLAB, we simulate a convolutional neural network 20 (CNN, Fig.”
MethodA Principled Design of Image Representation: Towards Forensic Tasks
Qi, S., Zhang, Y., Wang, C., Zhou, J., & Cao, X. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3204971 · Cited by 35
“Let us introduce the convolution theorem of the Fourier transform, such that the spatial-domain convolution in (15) can be converted to a frequency-domain product as [41]:”
BackgroundKalmia: A Heterogeneous QoS-aware Scheduling Framework for DNN Tasks on Edge Servers
Fu, Z., Ren, J., Zhang, D., Zhou, Y., & Zhang, Y. (2022) · IEEE Conference on Computer Communications (INFOCOM) · doi:10.1109/infocom48880.2022.9796661 · Cited by 30
“A common way to improve system throughput is to combine some DNN tasks into a batch [8].”
BackgroundSWIM: selective write-verify for computing-in-memory neural accelerators
Yan, Z., Hu, X. S., & Shi, Y. (2022) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3489517.3530459 · Cited by 28
“Non-volatile Computing-in-Memory (nvCiM) DNN accelerators [7] offer a great opportunity to edge applications by reducing data movement with an in-situ weight data access scheme [9].”
Wang, F., Cai, S., & Lau, V. K. N. (2022) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2022.3221850 · Cited by 34
“For example, the DNN task is computed on a layer-by-layer basis with strong sequential dependencies [1], [2], [3].”
BackgroundMethodAsynchronous Spiking Neurons, the Natural Key to Exploit Temporal Sparsity
Yousefzadeh, A., Serrano‐Gotarredona, T., Linares-Barranco, B., Khoei, M. A., Hosseini, S., Holanda, P., Leroux, S., Moreira, O., Tapson, J. C., Dhoedt, B., et al. (2019) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2019.2951121 · Cited by 38
“However, a certain dedicated accelerator can dramatically optimize deep ANN workloads, for instance by data reuse strategies and more efficient implementations of the computational parallelism [1].”
BackgroundSpecial Session: Approximation and Fault Resiliency of DNN Accelerators
Ahmadilivani, M. H., Barbareschi, M., Barone, S., Bosio, A., Daneshtalab, M., Torca, S. D., Gavarini, G., Jenihhin, M., Raik, J., Ruospo, A., et al. (2023) · IEEE VLSI Test Symposium (VTS) · doi:10.1109/vts56346.2023.10140043 · Cited by 28
“, healthcare, transportation, security, IoT and edge applications [1].”
BackgroundImproving Automated Machine-Learning Systems through Green AI
Castellanos–Nieves, D., & García-Forte, L. (2023) · Applied Sciences · doi:10.3390/app132011583 · Cited by 39
Pantheon: Preemptible Multi-DNN Inference on Mobile Edge GPUs
Han, L., Zhou, Z., & Li, Z. (2024) · International Conference on Mobile Systems, Applications, and Services · doi:10.1145/3643832.3661878 · Cited by 46
“The redundancy in DNNs allows for skipping certain operations with minimal impact on model accuracy [45, 59].”
BackgroundStructural nonlinear boundary condition identification using a hybrid physics data-driven approach
Luo, L., Sun, L.-M., Li, Y., & Xia, Y. (2024) · Nonlinear Dynamics · doi:10.1007/s11071-024-10614-x · Cited by 27
“The multilayer perceptron (MLP) is the foundation for numerous modern artificial intelligence applications, and it is still widely used nowadays [55].”
Bhandary, R., & Ghosh, B. K. (2025) · Journal of Risk and Financial Management · doi:10.3390/jrfm18010023 · Cited by 27
“In addition, the deep neural network is a fast-growing machine learning method because of its superior performance (Sze et al., 2017).”
MAX2: An ReRAM-Based Neural Network Accelerator That Maximizes Data Reuse and Area Utilization
Mao, M., Peng, X., Liu, R., Li, J., Yu, S., & Chakrabarti, C. (2019) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2019.2908937 · Cited by 27
“These networks achieve very high accuracy but at the price of large computational complexity [5].”
BackgroundDa-Zhong, M., Wang, J., Sun, Q., & Hu, X. (2019) · IEEE Access · doi:10.1109/access.2019.2908015 · Cited by 28
“Deep neural networks (DNNs) have special advantages in fault diagnosis due to the performance in processing vast information and classification [13]–[18].”
BackgroundCANN: Curable Approximations for High-Performance Deep Neural Network Accelerators
Hanif, M. A., Khalid, F., & Shafique, M. (2019) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3316781.3317787 · Cited by 44
“A more detailed overview of the neural networks and DNN hardware accelerators can be found in [18].”
BackgroundCompressing Deep Neural Networks With Sparse Matrix Factorization
Wu, K., Guo, Y., & Zhang, C. (2019) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2019.2946636 · Cited by 34
“Though undesirable in real-world applications [9], it is broadly believed that their success comes at the cost of overparameterization and, consequently, large computational complexity and memory complexity [10].”
BackgroundMachine learning methods for digital holography and diffractive optics
Аркадьевич, Ч. П., Evtikhiev, N. N., Krasnov, V. V., Rodin, V. G., Rymov, D. A., & Starikov, R. S. (2020) · Procedia Computer Science · doi:10.1016/j.procs.2020.02.243 · Cited by 28
MC2RAM: Markov Chain Monte Carlo Sampling in SRAM for Fast Bayesian Inference
Shukla, P., Shylendra, A., Tulabandhula, T., & Trivedi, A. R. (2020) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas45731.2020.9180701 · Cited by 28
Low Power Tiny Binary Neural Network with improved accuracy in Human Recognition Systems
Vita, A. D., Pau, D. P., Benedetto, L. D., Rubino, A., Pétrot, F., & Licciardo, G. D. (2020) · Euromicro Conference on Digital System Design (DSD) · doi:10.1109/dsd51259.2020.00057 · Cited by 30
Li, Y., Louri, A., & Karanth, A. (2021) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2021.3139015 · Cited by 29
“By contrast, the communication involved in CNN inference has several specific features such as nonuniform bandwidth demand between different chiplets, and recurrence of a few communication patterns [1], [2], [20], [21].”
BackgroundInfluentialKim, S., Shin, D. Y., Kim, T.-K., Lee, S., Hyun, J. K., & Park, S. (2022) · Sensors · doi:10.3390/s22020680 · Cited by 33
“As the dimension of the feature vector increases, the computational overhead increases; moreover, each extracted feature can have a different dimension [48,49].”
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.