Eyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks
The DOI matches its Crossref record; 38 of the 42 references it cites were found in a registry or an index.
Abstract
Eyeriss is an accelerator for state-of-the-art deep convolutional neural networks (CNNs). It optimizes for the energy efficiency of the entire system, including the accelerator chip and off-chip DRAM, for various CNN shapes by reconfiguring the architecture. CNNs are widely used in modern AI systems but also bring challenges on throughput and energy efficiency to the underlying hardware. This is because its computation requires a large amount of data, creating significant data movement from on-chip and off-chip that is more energy-consuming than computation. Minimizing data movement energy cost for any CNN shape, therefore, is the key to high throughput and energy efficiency. Eyeriss achieves these goals by using a proposed processing dataflow, called row stationary (RS), on a spatial architecture with 168 processing elements. RS dataflow reconfigures the computation mapping of a given shape, which optimizes energy efficiency by maximally reusing data locally to reduce expensive data movement, such as DRAM accesses. Compression and data gating are also applied to further improve energy efficiency. Eyeriss processes the convolutional layers at 35 frames/s and 0.0029 DRAM access/multiply and accumulation (MAC) for AlexNet at 278 mW (batch size N = 4), and 0.7 frames/s and 0.0035 DRAM access/MAC for VGG-16 at 236 mW (N = 3).
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
“such scaling will enable NeuRRAM to deliver higher energy and area efficiency than today’s state-of-the-art edge inference accelerators58–61.”
From A compute-in-memory chip based on resistive random-access memory (2022) · cites this paper as [58]
In the citing paper
“Some studies proposed to implement the argmax() function via hardware128–148, which could be beneficial to reduce the total transistor count and power consumption while at the same time increasing the throughput.”
From Hardware implementation of memristor-based artificial neural networks (2024) · cites this paper as [131]
In the citing paper
“Based on the analysis of the research, CoNNA is able to speed up at least 3.08 times and up to 14.1 times faster than the MIT Eyeriss Accelerator [20].”
From Embedded Intelligence on FPGA: Survey, Applications and Challenges (2021) · cites this paper as [20]
In the citing paper
“massively parallel processing architectures with distributed memory architectures, such as graphic processing units (GPUs) but also increasingly specialized chips for NN inference (e.g., [17]), have shown to be very energy efficient and capable of impressive computational throughput when fully utilized by concurrent algorithms [18].”
From An Introduction to Deep Learning for the Physical Layer (2017) · cites this paper as [17]
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 1×
“Modern CNNs achieve their superior accuracy by building a very deep hierarchy of layers [2]–[5], [7], which transform the input image data into highly abstract representations called feature maps (fmaps).”
BackgroundVery 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 7×
“We benchmark the chip performance using two publicly available and widely used CNNs: AlexNet [2] and VGG-16 [3].”
MethodInfluentialImageNet 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
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 1×
“[25] demonstrate a CNN processor and only report the theoretical peak throughput along with the power measured on a CNN for the MNIST data set [26], which has storage and computation requirements that are orders of magnitude lower than the state-of-the-art CNNs.”
MethodGoing 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 2×
“We have demonstrated an 1000-class image classification task [27] using this system, and a live demo can be found in [29].”
BackgroundMethodRich 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
Mastering the game of Go with deep neural networks and tree search
Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., et al. (2016) · Nature · doi:10.1038/nature16961 · Cited by 15,612 · cited 1×
“D EEP learning using convolutional neural networks (CNNs) [1] has achieved unprecedented accuracy on many modern AI applications [2]–[9].”
Background(2010) · International Conference on Machine Learning · Cited by 13,995
Research works printed but not linked: 2
- han, 2015, Learning both weights and connections for efficient neural network, Proc Adv Neural Inf Process Syst, 28, 1135
- chen, 2016, An 1000-Class Image Classification Task Performed by the Eyeriss-Integrated Deep Learning System
Cited by
Items 501–600 of 1,001
Evaluating and analyzing the energy efficiency of CNN inference on high‐performance GPU
Yao, C., Liu, W., Tang, W., Guo, J., Hu, S., Lu, Y., & Jiang, W. (2020) · Concurrency and Computation Practice and Experience · doi:10.1002/cpe.6064 · Cited by 49
Hsu, O., Strange, M., Sharma, R., Won, J., Olukotun, K., Emer, J., Horowitz, M., & Kjølstad, F. (2023) · International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) · doi:10.1145/3582016.3582051 · Cited by 61
“Prior work popularizes the separation of algorithm and schedule in both software [8, 46, 49, 56] and hardware [9].”
MethodSelf-citationChoi, K.-B., Woo, S. Y., Kang, W.-M., Lee, S., Kim, C.-H., Bae, J., Lim, S., & Lee, J. (2018) · Frontiers in Neuroscience · doi:10.3389/fnins.2018.00704 · Cited by 45
“However, software-based DNNs still have energy efficiency issues due to the energy consumed by the synchronous information processing and data movement between the processor and memory such as static random access memory (SRAM) and dynamic random access memory (DRAM) (Chen et al., 2017).”
BackgroundComputing-in-Memory with SRAM and RRAM for Binary Neural Networks
Sun, X., Liu, R., Peng, X., & Yu, S. (2018) · IEEE International Conference on Solid-State and Integrated Circuit Technology · doi:10.1109/icsict.2018.8565811 · Cited by 47
“multiplier and adders in the processing element (PE) [6, 7], as shown in Fig.”
Mixed-Signal Neuromorphic Inference Accelerators: Recent Results and Future Prospects
Bavandpour, M., Mahmoodi, M. R., Nili, H., Bayat, F. M., Prezioso, M., Vincent, A., Strukov, D. B., & Likharev, K. K. (2018) · 2018 IEEE International Electron Devices Meeting (IEDM) · doi:10.1109/iedm.2018.8614659 · Cited by 39
“Though the vast majority of demonstrated accelerators from industry [1-3] and academia [4, 5] are digital, the most natural approaches, however, are based on analog and mixed-signal circuits [6-13].”
BackgroundTaghavi, M., & Shoaran, M. (2019) · International IEEE/EMBS Conference on Neural Engineering · doi:10.1109/ner.2019.8716983 · Cited by 40
An 8 Bit 12.4 TOPS/W Phase-Domain MAC Circuit for Energy-Constrained Deep Learning Accelerators
Toyama, Y., Yoshioka, K., Ban, K., Maya, S., Sai, A., & Onizuka, K. (2019) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2019.2926649 · Cited by 38
“However, to the authors’ best knowledge, although many DLA system-level analyses targeting CNNs exist [16]–[19], there is little coverage of extremely energy-efficient DLAs processing timeseries data in the literature.”
BackgroundAutomated design of error-resilient and hardware-efficient deep neural networks
Schorn, C., Elsken, T., Vogel, S., Runge, A., Guntoro, A., & Ascheid, G. (2020) · Neural Computing and Applications · doi:10.1007/s00521-020-04969-6 · Cited by 48
Adam, K., Mohamed, I. I., & Ibrahim, Y. (2021) · IEEE Access · doi:10.1109/access.2021.3076716 · Cited by 49
Post-training Quantization with Multiple Points: Mixed Precision without Mixed Precision
Liu, X., Ye, M., Zhou, D., & Liu, Q. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i10.17054 · Cited by 55
Sense: Model-Hardware Codesign for Accelerating Sparse CNNs on Systolic Arrays
Sun, W., Liu, D., Zou, Z., Sun, W., Chen, S., & Kang, Y. (2023) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2023.3241933 · Cited by 42
“For example, Eyeriss [9] applies row stationary dataflow to minimize movement consumption; tensor processing unit (TPU) [10] revives systolic array to realize high-energy efficiency and throughput with structured dataflow and low bandwidth; Thinker [11] supports bit-width adaptive computing and on-demand array partiti…”
MethodInter-Thread Communication in Multithreaded, Reconfigurable Coarse-Grain Arrays
Voitsechov, D., Port, O., & Etsion, Y. (2018) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro.2018.00013 · Cited by 41
Neshatpour, K., Behnia, F., Homayoun, H., & Sasan, A. (2018) · Design, Automation, and Test in Europe · doi:10.23919/date.2018.8342068 · Cited by 40
“Computation of CONV layer in popular CNNs accounts for more than 90% of the overall operations and requires a large amount of data movement and memory operations [7].”
BackgroundEverson, L. R., Liu, M., Pande, N., & Kim, C. H. (2019) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2019.2914361 · Cited by 39
SparCE: Sparsity Aware General-Purpose Core Extensions to Accelerate Deep Neural Networks
Sen, S., Jain, S., Venkataramani, S., & Raghunathan, A. (2018) · IEEE Transactions on Computers · doi:10.1109/tc.2018.2879434 · Cited by 40
“incorporate a variety of compression techniques and zeroskipping schemes to reduce storage requirement and avoid redundant multiplications in accelerators [18], [26], [27],”
BackgroundMethodInfluentialA Training-Efficient Hybrid-Structured Deep Neural Network With Reconfigurable Memristive Synapses
Bai, K., An, Q., Liu, L., & Yi, Y. (2019) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2019.2942267 · Cited by 40
A dynamic CNN pruning method based on matrix similarity
Shao, M., Dai, J., Kuang, J., & Meng, D. (2020) · Signal Image and Video Processing · doi:10.1007/s11760-020-01760-x · Cited by 42
Efficient Scheduling of Irregular Network Structures on CNN Accelerators
Zheng, S., Zhang, X., Ou, D., Tang, S., Liu, L., Wei, S., & Yin, S. (2020) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2020.3012215 · Cited by 45
“In the previous experiments, the size of the PE array and the bandwidth provided are fixed, which is of the same magnitude with state-of-the-art designs (Eyeriss: 67.2GOP and Envision [10]: 102GOPS).”
BackgroundMethodInfluentialA Two-way SRAM Array based Accelerator for Deep Neural Network On-chip Training
Jiang, H., Huang, S., Peng, X., Su, J.-W., Chou, Y.-C., Huang, W.-H., Liu, T.-W., Liu, R., Chang, M., & Yu, S. (2020) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac18072.2020.9218524 · Cited by 44
“However, the memory wall problem still remains in most of the accelerators such as Eyeriss [1] and TPU [2] where the parameters are stored in embedded memory for data reuse and the computation is still performed at the digital multiply-and-cumulate (MAC) units.”
BackgroundBrain-inspired computing systems: a systematic literature review
Zolfagharinejad, M., Alegre-Ibarra, U., Chen, T., Kinge, S., & van der Wiel, W. G. (2024) · The European Physical Journal B · doi:10.1140/epjb/s10051-024-00703-6 · Cited by 39
Tong, J., Itagi, A., Chatarasi, P., & Krishna, T. (2024) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1109/isca59077.2024.00024 · Cited by 59
“…dataflows ( NEST ) Xilinx DPU [51], Gemmini [21] no reorder linear reduction parallel logarithmic reduction ( BIRRD ) SIMBA [47], Eyeriss [13] no reorder load imbalance across PE pick load-balance dataflows ( NEST ) Eyeriss v2 [15], SARA [44] off-chip high latency of moving data off-chip on-chip…”
Self-citationKim, K., Park, K., Yoon, K., Moon, H., Pyeon, H., Kim, J., & Han, S. (2025) · Journal of Power Sources · doi:10.1016/j.jpowsour.2025.236169 · Cited by 37
Adaptive weight compression for memory-efficient neural networks
Ko, J. H., Kim, D.-H., Na, T., Kung, J., & Mukhopadhyay, S. (2017) · Design, Automation, and Test in Europe · doi:10.23919/date.2017.7926982 · Cited by 37
Instruction driven cross-layer CNN accelerator with winograd transformation on FPGA
Yu, J., Hu, Y., Ning, X., Qiu, J., Guo, K., Wang, Y., & Yang, H. (2017) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/fpt.2017.8280147 · Cited by 42
“[3], [10] that utilize off-chip memory discuss the scheduling or data arrangement strategy to overcome the bandwidth limitation between the external and the onchip memory.”
BackgroundHeterogeneous Mixed-Signal Monolithic 3-D In-Memory Computing Using Resistive RAM
Murali, G., Sun, X., Yu, S., & Lim, S. K. (2020) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2020.3042411 · Cited by 43
Capstan: A Vector RDA for Sparsity
Rucker, A., Vilim, M., Zhao, T. Y., Zhang, Y., Prabhakar, R., & Olukotun, K. (2021) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1145/3466752.3480047 · Cited by 44
“Cnvlutin [3] skips zeros in CNN activations, and Eyeriss [9] and EIE [23] skip zeros in activations and weights using CSC. Capstan, on the other hand, has generic sparse iteration primitives, flexible inner-and outer-loop parallelism, and automatic pipelining.”
sEMG-Based Continuous Estimation of Finger Kinematics via Large-Scale Temporal Convolutional Network
Chen, C., Guo, W., Ma, C., Yang, Y., Wang, Z., & Lin, C. (2021) · Applied Sciences · doi:10.3390/app11104678 · Cited by 38
“However, the convolution structure is easy to accelerate in hardware and thus can provide better real-time performance [35], which was demonstrated in our previous work [36].”
BackgroundMemory-Efficient CNN Accelerator Based on Interlayer Feature Map Compression
Shao, Z., Chen, X., Du, L., Chen, L., Du, Y., Zhuang, W., Wei, H., Xie, C., & Wang, Z. (2021) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3120312 · Cited by 45
EAIS: Energy-aware adaptive scheduling for CNN inference on high-performance GPUs
Yao, C., Liu, W., Tang, W., & Hu, S. (2022) · Future Generation Computer Systems · doi:10.1016/j.future.2022.01.004 · Cited by 45
Optimally Scheduling CNN Convolutions for Efficient Memory AccessPreprint
Stoutchinin, A., Conti, F., & Benini, L. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1902.01492 · Cited by 57
VEGETA: Vertically-Integrated Extensions for Sparse/Dense GEMM Tile Acceleration on CPUs
Jeong, G., Damani, S., Bambhaniya, A. R., Qin, E., Hughes, C. J., Subramoney, S., Kim, H., & Krishna, T. (2023) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca56546.2023.10071058 · Cited by 53
Self-citationTileFlow: A Framework for Modeling Fusion Dataflow via Tree-based Analysis
Zheng, S., Chen, S., Gao, S., Jia, L., Sun, G., Wang, R., & Liang, Y. E. (2023) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1145/3613424.3623792 · Cited by 69
“Prior works such as GCNAX [33], OMEGA [14], Gamma [71], Spada [35], and Flexagon [43] support sparse workloads through customized PE arrays or caches.”
MethodHardware-Aware DNN Compression via Diverse Pruning and Mixed-Precision Quantization
Balaskas, K., Καρατζάς, Α., Sad, C., Siozios, K., Anagnostopoulos, I., Zervakis, G., & Henkel, J. (2024) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2023.3346944 · Cited by 51
“Energy gains are reported with respect to the baseline 8-bit Eyeriss-based [2] accelerator (details regarding the experimental setup and the energy evaluation are found in Section 5).”
BackgroundMethodInfluentialOpportunities for Machine Learning in Electronic Design Automation
Beerel, P. A., & Pedram, M. (2018) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas.2018.8351731 · Cited by 42
A Resource-Efficient Multiplierless Systolic Array Architecture for Convolutions in Deep Networks
Parmar, Y., & Sridharan, K. (2019) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2019.2907974 · Cited by 36
“It is desirable to have a hardware-efficient architecture for CNN-based mobile applications [5] keeping also accuracy in view.”
BackgroundSong, Z., Liu, Z., & Wang, D. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11334 · Cited by 41
Exploiting Retraining-Based Mixed-Precision Quantization for Low-Cost DNN Accelerator Design
Kim, N., Shin, D., Choi, W., Kim, G., & Park, J. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.3008996 · Cited by 39
Real-Time SSDLite Object Detection on FPGA
Kim, S., Na, S., Kong, B. Y., Choi, J., & Park, I. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2021.3064639 · Cited by 42
Yang, J., Kong, Y., Zhang, Z., Liu, Z., Zhou, J. L., Wang, Y., Liu, Y., Guo, C., Hu, T., Li, C., et al. (2021) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2021.3095232 · Cited by 42
Heterogeneous Systolic Array Architecture for Compact CNNs Hardware Accelerators
Xu, R., Ma, S., Wang, Y., Guo, Y., Li, D., & Qiao, Y. (2021) · IEEE Transactions on Parallel and Distributed Systems · doi:10.1109/tpds.2021.3129647 · Cited by 43
“Systolic arrays have become the main trend in hardware CNNs accelerators deployed in both edge devices [11], [12], [16], [26], [27] and servers used in data-centers [25], [28], [29], [30].”
BackgroundDeepOBS: A Deep Learning Optimizer Benchmark SuitePreprint
Schneider, F., Balles, L., & Hennig, P. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1903.05499 · Cited by 87
“…mainstream, research on aspects like architectures (Graves et al., 2014; He et al., 2016; Szegedy et al., 2017; Vaswani et al., 2017; Sabour et al., 2017) and hardware (Ovtcharov et al., 2015; Chen et al., 2016; Reagen et al., 2016; Jouppi, 2016) has exploded, and helped professionalize the field.”
BackgroundKim, S., Kim, S., Hong, S., Kim, S., Han, D., Choi, J., & Yoo, H. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3330483 · Cited by 35
“To generate one output in address #0 of the output buffer by using row-stationary data reuse [14], three input data (addresses #0, #1, and #2 in the input buffer) and three weight data (addresses #0, #1, and #2 in the weight buffer) are required (at 3 × 3 kernel convolution case).”
MethodLerGAN: A Zero-Free, Low Data Movement and PIM-Based GAN Architecture
Mao, H., Song, M., Li, T., Dai, Y., & Shu, J. (2018) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1109/micro.2018.00060 · Cited by 42
Recurrent Residual Module for Fast Inference in Videos
Pan, B., Lin, W., Fang, X., Huang, C., Zhou, B., & Lu, C. (2018) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2018.00166 · Cited by 40
PreVIous: A Methodology for Prediction of Visual Inference Performance on IoT Devices
Velasco-Montero, D., Fernández‐Berni, J., Carmona-Galan, R., & Rodriguez-Vazquez, A. (2020) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.2981684 · Cited by 43
“ciency of CNN accelerators such as Eyeriss [45], an energy estimation methodology was proposed [46], [47].”
MethodFPGA Implementation of Quantized Convolutional Neural Networks
Zhang, Q., Cao, J., Zhang, Y., Zhang, S., Zhang, Q., & Yu, D. (2019) · IEEE International Conference on Communication Technology (ICCT) · doi:10.1109/icct46805.2019.8947168 · Cited by 39
“Another important research direction is to use low-precision weights and activation to implement inference, sometimes requiring back-propagation training [9]-[12].”
BackgroundDeep Learning Acceleration with Neuron-to-Memory Transformation
Imani, M., Samragh, M., Kim, Y., Gupta, S., Koushanfar, F., & Rosing, T. (2020) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca47549.2020.00011 · Cited by 40
RRAM-DNN: An RRAM and Model-Compression Empowered All-Weights-On-Chip DNN Accelerator
Li, Z., Wang, Z., Xu, L., Dong, Q., Liu, B., Su, C.-I., Chu, W.-T., Tsou, G., Chih, Y.-D., Chang, T.-Y. J., et al. (2020) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2020.3045369 · Cited by 40
Special Session: Reliability Analysis for AI/ML Hardware
Kundu, S., Basu, K., Sadi, M., Titirsha, T., Song, S., Das, A. K., & Guin, U. (2021) · IEEE VLSI Test Symposium (VTS) · doi:10.1109/vts50974.2021.9441050 · Cited by 42
“Training Phase CPU Inference Phase [1] Trained weights and connections [2] Acceptable fault rate for PEs in the accelerator for inference (FRmax_non.”
BackgroundMethodInfluentialDFSynthesizer: Dataflow-based Synthesis of Spiking Neural Networks to Neuromorphic Hardware
Song, S., Chong, H., Balaji, A., Das, A. K., Shackleford, J., & Kandasamy, N. (2022) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3479156 · Cited by 40
“Energy consumed by neuromorphic hardware can be several orders of magnitude lower than a conventional machine-learning accelerator such as Eyeriss [26].”
BackgroundHiMap: Fast and Scalable High-Quality Mapping on CGRA via Hierarchical Abstraction
Wijerathne, D., Li, Z., Pathania, A., Mitra, T., & Thiele, L. (2021) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2021.3132551 · Cited by 37
“Many spatial dataflow accelerators also use systolic arrays [23]–[25].”
BackgroundCao, T., Liu, C., Wang, W., Zhang, T., Lee, H. K., Li, M. H., Song, W., Chen, Z. X., Zhuo, V. Y.-Q., Wang, N., et al. (2022) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2022.3214334 · Cited by 34
Reconfigurability, Why It Matters in AI Tasks Processing: A Survey of Reconfigurable AI Chips
Wei, S., Lin, X., Tu, F., Wang, Y., Liu, L., & Yin, S. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3228860 · Cited by 40
RAMAN: A Reconfigurable and Sparse tinyML Accelerator for Inference on Edge
Krishna, A., Nudurupati, S. R., Chandana, D., Dwivedi, P., André, V. S., Mehendale, M. M., & Thakur, C. S. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3386832 · Cited by 50
“A customized hardware design for neural networks provides an opportunity to optimize dataflow, memory access, and exploit network sparsity to overcome edge computing bottlenecks.”
BackgroundMECLA: Memory-Compute-Efficient LLM Accelerator with Scaling Sub-matrix Partition
Qin, Y., Wang, Y., Zhao, Z., Yang, X., Zhou, Y., Wei, S., Hu, Y., & Yin, S. (2024) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1109/isca59077.2024.00079 · Cited by 49
Boosting Occluded Image Classification via Subspace Decomposition-Based Estimation of Deep Features
Cen, F., & Wang, G. (2019) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2019.2931067 · Cited by 36
“With the development of deep neural network accelerator chips [15], the CNN-based feature extraction, used to be conducted at the cloud or server end due to its high computational complexity, are gradually shifting to the terminal or mobile end.”
MethodAutomatic Generation of Multi-Precision Multi-Arithmetic CNN Accelerators for FPGAs
Zhao, Y., Gao, X., Guo, X., Liu, J., Wang, E., Mullins, R., Cheung, P. Y. K., Constantinides, G. A., & Xu, C. (2019) · International Conference on Field-Programmable Technology (FPT) · doi:10.1109/icfpt47387.2019.00014 · Cited by 42
Putra, R. V. W., Hanif, M. A., & Shafique, M. (2020) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac18072.2020.9218672 · Cited by 43
Oxide-based filamentary RRAM for deep learning
Zhang, Y., Huang, P. C., Gao, B., Kang, J., & Wu, H. (2020) · Journal of Physics D Applied Physics · doi:10.1088/1361-6463/abc5e7 · Cited by 39
Efficiency Versus Accuracy: A Review of Design Techniques for DNN Hardware Accelerators
Latotzke, C., & Gemmeke, T. (2021) · IEEE Access · doi:10.1109/access.2021.3050670 · Cited by 39
“A good example for the effect of RLC compression combined with data flow optimization is [49]–[51].”
BackgroundInfluentialDistributed Intelligence in Wireless Networks
Liu, X., Yu, J., Liu, Y., Gao, Y., Mahmoodi, T., Lambotharan, S., & Tsang, D. H. K. (2023) · IEEE Open Journal of the Communications Society · doi:10.1109/ojcoms.2023.3265425 · Cited by 48
“…the convincing reasons to apply DL to deal with such concern is that the trained DL methods can pass through parallel distributed memory architectures, such as the graphic processing units and specialized computation chips that can demonstrate fast and energy-efficient computational ability [200].”
MethodStream Processing Dual-Track CGRA for Object Inference
Fan, X., Wu, D., Cao, W., Luk, W. W., & Wang, L. (2018) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2018.2797600 · Cited by 38
“As an ASIC implementation, [9] achieves a better energy efficiency than SDT-CGRA, due to its highly customized (for CNN) architecture and memory system.”
BackgroundMethodInfluentialSecure XOR-CIM Engine: Compute-In-Memory SRAM Architecture With Embedded XOR Encryption
Huang, S., Jiang, H., Peng, X., Li, W., & Yu, S. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2021.3120296 · Cited by 36
DTQAtten: Leveraging Dynamic Token-based Quantization for Efficient Attention Architecture
Yang, T., Li, D., Song, Z., Zhao, Y., Liu, F., Wang, Z., He, Z., & Jiang, L. (2022) · Design, Automation, and Test in Europe · doi:10.23919/date54114.2022.9774692 · Cited by 43
“Where the implementations of Eyeriss-based design without model compression are normalized to 1.”
MethodYang, T., Ma, F., Li, X., Liu, F., Zhao, Y., He, Z., & Jiang, L. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3181541 · Cited by 33
“For hardware inference performance, we compare our design (with DTATrans models, mapped on the VSSA using our optimization strategy) with two kinds of accelerators: 1) NN-based accelerators, including: a) Eyeriss [32] (with original Transformer models); b) Systolic-base (with DTQAtten models, but mapped on VSSA withou…”
MethodInfluentialNeuroCARE: A generic neuromorphic edge computing framework for healthcare applications
Tian, F., Yang, J., Zhao, S., & Sawan, M. (2023) · Frontiers in Neuroscience · doi:10.3389/fnins.2023.1093865 · Cited by 39
“The architecture proposed in Eyeriss (Chen et al., 2017) is adopted in our work.”
MethodInfluentialWang, Z., & Verma, N. (2017) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2017.2703880 · Cited by 33
Huang, H., Ni, L., Wang, K., Wang, Y., & Yu, H. (2017) · IEEE Transactions on Nanotechnology · doi:10.1109/tnano.2017.2732698 · Cited by 40
Hardware-Aware Softmax Approximation for Deep Neural Networks
Geng, X., Lin, J., Zhao, B., Kong, A., Aly, M. M. S., & Chandrasekhar, V. (2019) · Lecture notes in computer science · doi:10.1007/978-3-030-20870-7_7 · Cited by 40
Unrolling Ternary Neural Networks
Tridgell, S., Kumm, M., Hardieck, M., Boland, D., Moss, D. J. M., Zipf, P., & Leong, P. H. W. (2019) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3359983 · Cited by 40
PIMBALL: Binary Neural Networks in Spintronic Memory
Resch, S., Khatamifard, S. K., Chowdhury, Z. I., Zabihi, M., Zhao, Z., Wang, J., Sapatnekar, S. S., & Karpuzcu, U. R. (2019) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3357250 · Cited by 38
“umerous NN accelerators have been proposed for forward propagation on traditional compute substrates. For example, DaDianNao [5] uses a multi-chip system to implement high-precision networks. Eyeriss [6] develops a data flow to improve the energy efficiency on a spatial architecture. Tetris [16] and Neurocube [23] uti…”
MethodCross-Stack Workload Characterization of Deep Recommendation Systems
Hsia, S., Gupta, U., Wilkening, M., Wu, C.-J., Wei, G.-Y., & Brooks, D. M. (2020) · IEEE International Symposium on Workload Characterization · doi:10.1109/iiswc50251.2020.00024 · Cited by 47
Deep learning‐based pilot‐assisted channel state estimator for OFDM systems
Essai, M. H. (2020) · IET Communications · doi:10.1049/cmu2.12051 · Cited by 35
RSNN: A Software/Hardware Co-Optimized Framework for Sparse Convolutional Neural Networks on FPGAs
You, W., & Wu, C. (2020) · IEEE Access · doi:10.1109/access.2020.3047144 · Cited by 34
Automated brain tumor classification using various deep learning models: a comparative study
Abbood, A. A., Shallal, Q. M., & Fadhel, M. A. (2021) · Indonesian Journal of Electrical Engineering and Computer Science · doi:10.11591/ijeecs.v22.i1.pp252-259 · Cited by 49
“[26], CNN's method was applied for facial classification based on edge detection.”
BackgroundMethodA survey of in-spin transfer torque MRAM computing
Cai, H., Liu, B., Chen, J., Naviner, L., Zhou, Y., Wang, Z., & Yang, J. (2021) · Science China Information Sciences · doi:10.1007/s11432-021-3220-0 · Cited by 33
Park, J., Lee, S., & Jeon, D. (2021) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2021.3103603 · Cited by 41
Deep Neural Network: An Alternative to Traditional Channel Estimators in Massive MIMO Systems
Melgar, A., de la Fuente, A., Carro‐Calvo, L., Barquero‐Pérez, Ó., & Morgado, E. (2022) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2022.3164888 · Cited by 45
Liao, S., Li, Z., Lin, X., Qiu, Q., Wang, Y., & Yuan, B. (2017) · International Conference on Computer Aided Design · doi:10.1109/iccad.2017.8203813 · Cited by 39
“In [7], [8], the strategies of pruning DNNs were used to develop the corresponding hardware since a large amount of energy and area can be saved after removing unnecessary weights or neurons.”
MethodCheng, X., Liu, D., Lu, J., Wei, L., Hu, A., Lei, J., Zou, Z., Zou, X., & Jiang, Q. (2022) · Microelectronics Journal · doi:10.1016/j.mejo.2022.105492 · Cited by 33
Zhang, H., Liu, J., Bai, J., Li, S., Luo, L., Wei, S., Wu, J., & Kang, W. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3199440 · Cited by 37
Dedicated FPGA Implementation of the Gaussian TinyYOLOv3 Accelerator
Ki, S., Park, J., & Kim, H. (2023) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2023.3289514 · Cited by 50
“However, GPUs exhibit drawbacks such as significant power consumption and a lack of cost-effectiveness, making the optimization of specific model implementations challenging [4], [6].”
BackgroundZhang, H., He, S., Lu, X., Guo, X., Wang, S., Du, Y., & Du, L. (2023) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2023.3301814 · Cited by 38
“However, quantized CNNs are dominated by parallel and memory-intensive integer multiply-and-accumulate (MAC) operations, which is a primary concern for battery-constrained edge processors [5], [6], [7].”
BackgroundEvent-triggered natural hazard monitoring with convolutional neural networks on the edge
Meyer, M., Farei-Campagna, T., Pasztor, A., Forno, R. D., Gsell, T., Faillettaz, J., Vieli, A., Weber, S., Beutel, J., & Thiele, L. (2019) · ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN) · doi:10.1145/3302506.3310390 · Cited by 45
“Artificial Neural Networks on Embedded Devices: Many studies focus on additional accelerators [4, 10, 18] for convolutional neural networks.”
BackgroundHan, X., Zhou, D., Wang, S., & Kimura, S. (2016) · IEEE International Conference on Computer Design: VLSI in Computers and Processors (ICCD) · doi:10.1109/iccd.2016.7753296 · Cited by 39
“Grey filters are composed of co-located filters. n co-located windows and n filters in one group determine one element in output feature map. m group of filters corresponds m co-located output elements”
BackgroundA RISC-V based hardware accelerator designed for Yolo object detection system
Zhang, G., Zhao, K., Wu, B., Sun, Y., Sun, L., & Liang, F. (2019) · IEEE International Conference of Intelligent Applied Systems on Engineering (ICIASE) · doi:10.1109/iciase45644.2019.9074051 · Cited by 36
“To increase the performance of accelerator, [5] deceases the movement of data from on-chip and off-chip by a reconfigurable architecture.”
BackgroundPutra, 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
“Layer Partitioning and Scheduling The layer partitioning and scheduling of convolutional operations in a CNN are typically devised for each layer of a network, hence most of the CNN accelerators process a network sequentially, one layer at a time [5]–[8].”
MethodAPNAS: Accuracy-and-Performance-Aware Neural Architecture Search for Neural Hardware Accelerators
Achararit, P., Hanif, M. A., Putra, R. V. W., Shafique, M., & Hara–Azumi, Y. (2020) · IEEE Access · doi:10.1109/access.2020.3022327 · Cited by 42
Sayal, A., Fathima, S., Nibhanupudi, S. S. T., & Kulkarni, J. P. (2020) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2020.3041502 · Cited by 37
Han, D., Lee, J., & Yoo, H. (2020) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2020.3042978 · Cited by 34
Accelerating Deep Neural Networks implementation: A survey
Dhouibi, M., Salem, A. K. B., Saidi, A., & Saoud, S. B. (2021) · IET Computers & Digital Techniques · doi:10.1049/cdt2.12016 · Cited by 35
“Several research works have focussed on building customised ASICs to accelerate DL model training and inference [44].”
BackgroundLin, W., Adetomi, A., & Arslan, T. (2021) · Electronics · doi:10.3390/electronics10172048 · Cited by 39
“References [20,34,36,40] are not included in Figure 1 because they lack the area data at the chip level.”
BackgroundMethodInfluentialMarvel: A Data-Centric Approach for Mapping Deep Learning Operators on Spatial Accelerators
Chatarasi, P., Kwon, H., Parashar, A., Pellauer, M., Krishna, T., & Sarkar, V. (2021) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3485137 · Cited by 41
“The MDC’s cost model precisely computes performance and energy, accounting for under-utilization, edge conditions, and data reuse or movement across time (via L1/L2 buffers [7]), space (via broadcast links [26]), and space-time (via neighboring links [9, 20]) without requiring explicit RTL/cycle-level simulations or a…”
BackgroundSelf-citationETA: 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
Efficient Algorithms for Device Placement of DNN Graph OperatorsPreprint
Tarnawski, J., Phanishayee, A., Devanur, N. R., Mahajan, D., & Nina-Paravecino, F. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.16423 · Cited by 84
Approximate Computing Methods for Embedded Machine Learning
Ibrahim, A., Osta, M., Alameh, M., Saleh, M., Chible, H., & Valle, M. (2018) · 2018 25th IEEE International Conference on Electronics, Circuits and Systems (ICECS) · doi:10.1109/icecs.2018.8617877 · Cited by 31
“Memory Storage Memory is a big challenge in embedded machine learning and deep neural networks today [7], [8].”
BackgroundReggiani, E., Pappalardo, A., Doblas, M., Moretó, M., Olivieri, M., Ünsal, O., & Cristal, A. (2023) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca56546.2023.10071076 · Cited by 54
“7 × to 24 × with respect to Eyeriss, and from 1 .”
BackgroundInfluentialCiMLoop: A Flexible, Accurate, and Fast Compute-In-Memory Modeling Tool
Andrulis, T., Emer, J., & Sze, V. (2024) · IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) · doi:10.1109/ispass61541.2024.00012 · Cited by 60
“CiMLoop models the CiM macro space and integrates CiM macros with system modeling tools [9]–[14].”
Self-citationMemristor Crossbar Tiles in a Flexible, General Purpose Neural Processor
Mountain, D. J., McLean, M. R., & Krieger, C. D. (2017) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2017.2767024 · Cited by 32
A Low-latency Sparse-Winograd Accelerator for Convolutional Neural Networks
Wang, H., Liu, W., Xu, T., Lin, J. T., & Wang, Z. (2019) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp.2019.8683512 · Cited by 30
“Hence, many works of CNN accelerators [6–8] have been proposed based on the Field Programmable Gate Arrays (FPGA) or Application Specific Integrated Circuit (ASIC) to achieve lower latency, less energy consumption, and higher flexibility.”
Background
Metadata
- Authors
- Chen, Y.-H., Krishna, T., Emer, J. S., & Sze, V.
- Affiliations
- Massachusetts Institute of Technology (US)
- Georgia Institute of Technology (US)
- Nvidia (United States) (US)
- Venue
- IEEE Journal of Solid-State Circuits
- Published
- 2017-01
- Volume
- 52
- Issue
- 1
- Pages
- 127-138
- ISSN
- 0018-9200 (ISSN-L)
- 1558-173X
- OpenAlex
- W2289252105
- Semantic Scholar
- ffdaa12ef011de9dbf43be46d45a3abcc8288965
- Cited by
- 3,247
- References
- 42
- Citations by year
- 2026: 218 · 2025: 313 · 2024: 294 · 2023: 373 · 2022: 365 · 2021: 450 · 2020: 427 · 2019: 354 · 2018: 302 · 2017: 118
- FWCI
- 182.12
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- IEEE Journal of Solid-State Circuits (publishedVersion)
- 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
- reconfigurable accelerators, deep convolutional neural networks, energy efficiency, spatial architecture, data movement reduction, data compression, processing elements, AlexNet, VGG16, convolutional layers
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.