Fused-layer CNN accelerators
The DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index.
Abstract
Deep convolutional neural networks (CNNs) are rapidly becoming the dominant approach to computer vision and a major component of many other pervasive machine learning tasks, such as speech recognition, natural language processing, and fraud detection. As a result, accelerators for efficiently evaluating CNNs are rapidly growing in popularity. The conventional approaches to designing such CNN accelerators is to focus on creating accelerators to iteratively process the CNN layers. However, by processing each layer to completion, the accelerator designs must use off-chip memory to store intermediate data between layers, because the intermediate data are too large to fit on chip. In this work, we observe that a previously unexplored dimension exists in the design space of CNN accelerators that focuses on the dataflow across convolutional layers. We find that we are able to fuse the processing of multiple CNN layers by modifying the order in which the input data are brought on chip, enabling caching of intermediate data between the evaluation of adjacent CNN layers. We demonstrate the effectiveness of our approach by constructing a fused-layer CNN accelerator for the first five convolutional layers of the VGGNet-E network and comparing it to the state-of-the-art accelerator implemented on a Xilinx Virtex-7 FPGA. We find that, by using 362KB of on-chip storage, our fused-layer accelerator minimizes off-chip feature map data transfer, reducing the total transfer by 95%, from 77MB down to 3.6MB per image.
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
“When the number of worker nodes increases from 2 to 4, MoDNN can speedup the DNN computation by 2.17-4.28×[130]”
From Convergence of Edge Computing and Deep Learning: A Comprehensive Survey (2020) · cites this paper as [130]
In the citing paper
“With regard to specific DNN structures, e.g., CNN, a finer grid partitioning can be applied to minimize communication, synchronization, and memory overhead [130].”
From Convergence of Edge Computing and Deep Learning: A Comprehensive Survey (2020) · cites this paper as [130]
Citations
Cites
Very 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 6×
“Table II shows the results of a similar comparison, but for the larger and newer VGGNet-E network [17].”
BackgroundMethodResultInfluentialImageNet 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 4×
“(We observe a similar pattern in other common CNN structures, such as AlexNet [8].”
BackgroundMethodInfluentialGoing 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 3×
“Large Scale Visual Recognition Challenge (ILSVRC) [15], uses five convolutional layers, while VGG [17] (winner of a portion of ILSVRC 2014) uses up to 17 convolutional layers.”
MethodOptimizing FPGA-based Accelerator Design for Deep Convolutional Neural Networks
Zhang, C., Li, P., Sun, G., Guan, Y., Xiao, B., & Cong, J. (2015) · Proceedings of the 2015 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays · doi:10.1145/2684746.2689060 · Cited by 2,158 · cited 10×
“The computational dominance of the convolutional layers, coupled with the need for larger and deeper networks, has sparked significant interest in the design and optimization of accelerator structures for these layers [19], [10].”
BackgroundMethodResultInfluentialDaDianNao: A Machine-Learning Supercomputer
Chen, Y., Luo, T., Liu, S., Zhang, S., He, L., Wang, J., Li, L., Chen, T., Xu, Z., Sun, N., et al. (2014) · 2014 47th Annual IEEE/ACM International Symposium on Microarchitecture · doi:10.1109/micro.2014.58 · Cited by 1,492 · cited 3×
“As a result, there has been significant interest in developing and adapting hardware accelerators for CNNs [7] such as GPUs [3], FPGAs [13], [19], [10], [1], and ASICs [2].”
BackgroundMethodTorch7: A Matlab-like Environment for Machine Learning
Collobert, R., Kavukcuoglu, K., & Farabet, C. (2011) · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · Cited by 1,186
cuDNN: Efficient Primitives for Deep Learning
Chetlur, S., Woolley, C., Vandermersch, P., Cohen, J., Tran, J. J., Catanzaro, B., & Shelhamer, E. (2014) · arXiv (Cornell University) · doi:10.48550/arxiv.1410.0759 · Cited by 943 · cited 2×
“As a result, there has been significant interest in developing and adapting hardware accelerators for CNNs [7] such as GPUs [3], FPGAs [13], [19], [10], [1], and ASICs [2].”
MethodA reconfigurable fabric for accelerating large-scale datacenter services
Putnam, A., Caulfield, A. M., Chung, E. S., Chiou, D., Constantinides, K., Demme, J., Esmaeilzadeh, H., Fowers, J., Gopal, G. P., Gray, J., et al. (2014) · 2014 ACM/IEEE 41st International Symposium on Computer Architecture (ISCA) · doi:10.1109/isca.2014.6853195 · Cited by 576
NeuFlow: A runtime reconfigurable dataflow processor for vision
Farabet, C., Martini, B., Corda, B., Akselrod, P., Culurciello, E., & LeCun, Y. (2011) · CVPR 2011 WORKSHOPS · doi:10.1109/cvprw.2011.5981829 · Cited by 369 · cited 1×
“In [6], [5], [16], systolic implementations were proposed.”
Cited by
Items 701–779 of 779
Liu, L. (2024) · Scalable Computing Practice and Experience · doi:10.12694/scpe.v25i4.2929
Basics and Research Status of Neural Network Processors
Yue, J. (2024) · Springer theses · doi:10.1007/978-981-97-3477-1_2
TeMCO: Tensor Memory Compiler Optimization across Tensor Decompositions in Deep Learning Inference
Song, S., Lee, J. M., Jeong, H., Kwon, H., Jeong, S., Lee, J., & Kim, H. (2024) · Proceedings of the International Conference on Parallel Processing · doi:10.1145/3673038.3673048
“To clarify our definition, activation layer fusion means the fusion of a lconv -activation- fconv sequence, not the fusion [3, 27] of a convolution-activation pair, nor the activation data compression [5, 14] in deep learning training.”
Performance Analysis of CNNs Deployed on Low Power Arquitectures
Pérez-Peña, A. M., Linares-Barranco, A., Morato, D. C., Barranco, F., & Romero, S. F. (2024) · Springer Proceedings in Materials · doi:10.1007/978-3-031-64106-0_63
A Hardware Architecture for the CBL Block based Pyramid-shaped Multilayer Sliding Window
Kim, K., & Moon, B. (2024) · The Journal of Korean Institute of Information Technology · doi:10.14801/jkiit.2024.22.9.89
Yi, D., Lee, S., Kim, J.-H., Kim, J., Ha, S., Chang, I., & Je, M. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2411.14733
Deep Learning Optimization and Acceleration
Jiang, B. (2022) · University of North Texas Digital Library (University of North Texas) · doi:10.12794/metadc1986716
Acceleration of CUDA Kernels through Fusion Measurements for AvgPool2D and ReLU
Falkenberg, A. (2025) · doi:10.5121/csit.2024.150213
“[1 ] [2] [3] They also need to adapt quickly to support new features of new hardware.”
Acceleration through Fusion of AvgPool2D and Silu Kernels
Falkenberg, A. (2025) · doi:10.5121/csit.2024.150210
“[1] [2] [3] They also need to adapt quickly to support new features of new hardware.”
SPONGE: A Scalable Heterogeneous Chiplet Accelerator Architecture for Autonomous Driving
Tang, C., Jiang, Y., Fu, W., Shi, H., Huang, P. S., Wang, L., Huang, M., Qiao, Y., Zhou, Y., & Guan, P. (2025) · International Symposium on Computer Applications and Information Systems (ISCAIS) · doi:10.1109/iscait64916.2025.11010368
“Due to the computational and memory-intensive nature of autonomous driving DNNs, numerous accelerators have been proposed to enhance the speed and efficiency of their computations [1][2][3][4][6][7][9] .”
Two Mapping Strategies for Autonomous Driving Accelerators
Tang, C., Jiang, Y., Fu, W., Shi, H., Huang, P., Wang, L., Huang, M., Qiao, Y., Zhou, Y., & Guan, P. (2025) · International Symposium on Computer Applications and Information Systems (ISCAIS) · doi:10.1109/iscait64916.2025.11010768
“By evaluating our optimizations on these widely used models, we ensure that the results are representative and applicable to a broad spectrum of DNN applications [5][6] .”
Multi-DOF Fusion: A Flexible Fusion Strategy for Reducing Redundancy in CNN Workloads
Chen, Y., Zhou, Z., Dai, S., Duan, X., & Han, J. S. (2025) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3716368.3735173
Cello: Co-Designing Schedule and Hybrid Implicit/Explicit Buffer for Complex Tensor Reuse
Garg, R., Pellauer, M., Rajamanickam, S., & Krishna, T. (2025) · IEEE International Parallel and Distributed Processing Symposium (IPDPS) · doi:10.1109/ipdps64566.2025.00046
“…[25], Timeloop [32], CoSA [18], GAMMA [21], Intersteller [43], TPUv1 [20], MAERI [26], Eyeriss [7], Eyeriss v2 [8] FusedCNN [3], FLAT [22], FlashAttention [9], X-layer [39], ISOSceles [44], TileFlow [45], OMEGA [14] Can reuse among adjacent ops only when there is no delayed dependency SET [6],…”
InfluentialMemory-Efficient Feature Merging for Residual Connections with Layer-Centric Tile Fusion
Zhang, H., He, J., Gui, Y., Peng, S., Huang, L., Yan, X., & Fan, Y. (2025) · Electronics · doi:10.3390/electronics14163269
Yu, R., Zou, C., Li, J., Gao, B., Tang, J., Qian, H., & Wu, H. (2025) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2025.3600366
“Addressing this issue, researchers have proposed optimization methods such as fine-grained pipelining in architectural design [8], [20].”
Modeling DRAM Access Based on Efficient Tiling in CNN Hardware Accelerators
Seyyedi, S., & Nasab, M. E. S. (2025) · Journal of Association of Electrical and Electronics Engineers · doi:10.61882/jiaeee.22.2.160
Towards an efficient dataflow-flexible accelerator by finding optimal dataflows of DNNs
Kim, H., Ha, W. R., Lee, Y., Lee, D.-J., Lee, J., Woo, D., Yoon, J. W., Lee, J., Kwon, Y., & Paek, Y. (2025) · Future generations computer systems · doi:10.1016/j.future.2025.108123
Design Consideration of Scalable and Adaptive FPGA-Based Convolutional Layer Accelerator
Peetakul, J., Nilwong, S., Chuayphan, N., Prapan, S., & Sukjai, O. (2025) · International Conference on Business and Industrial Research (ICBIR) · doi:10.1109/icbir65229.2025.11163238
“This design methodology, often referred to as a scale-up systolic array (SUSA) architecture, is widely adopted in many DNN accelerator implementations [9] [10].”
Peng, C., Huang, H., Huang, X., & Tang, Z. (2025) · Journal of Mechanical Science and Technology · doi:10.1007/s12206-025-0922-x
“At the same time, 1D-CNN has high computational efficiency and is suitable for training after large-scale data compression, which helps to accelerate the model training process [31].”
CIMR-V: An End-to-End SRAM-based CIM Accelerator with RISC-V for AI Edge DevicePreprint
and, Y.-C. G., Chang, T., Lin, C.-S., Chiou, B.-C., Lai, C.-M., Sheu, S.-S., Lo, W., & Chang, S.-C. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.22072
Ding, Y.-C., Chang, C.-Y., Tsai, H.-Y., Lin, C.-Y., Tu, H.-J., Su, Y.-C., Chang, K.-F., Hsieh, T.-H., Chen, W.-C., Chang, N.-S., et al. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3628249
“LF [17] reduces this to 227 MB, but real-time deployment on edge devices remains di ffi cult.”
Tang, C., Huang, Z., Wang, W., Feng, X., He, Y., Fan, S., Wang, K., Sun, W., & Liu, Y. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3622515
“However, such solutions are less e ff ective in scenarios dominated by frequent weight loading, as explained in [53].”
Hardware-Software Co-design for Improved Resource Utilization in DNN Accelerators
Yüzügüler, A. C. (2023) · Infoscience (Ecole Polytechnique Fédérale de Lausanne) · doi:10.5075/epfl-thesis-9884
Estimation, Profiling and Modeling of DNNs for Embedded Systems
Wess, M. (2025) · reposiTUm (TU Wien) · doi:10.34726/hss.2025.43436
An Analytical Cost Model for Fast Evaluation of Multiple Compute-Engine CNN AcceleratorsPreprint
Qararyah, F. M., Maleki, M. A., & Trancoso, P. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2503.07242
“A systematic accelerator design approach requires analyzing the bottlenecks and inefficiencies and exploring the design space to identify optimizations that mitigate them.”
Holistic Optimization of Embedded Computer Vision Systems
Buckler, M. (2019) · eCommons (Cornell University) · doi:10.7298/tcy7-vx76
Accelerating Data Movement at Different Granularities in Datacenters
Wang, S.-T. (2024) · eScholarship (California Digital Library)
Stahl, R., Müller-Gritschneder, D., & München, T. U. (2023)
Zhang, Y. (2024) · eScholarship (California Digital Library)
Efficient Processing of Compact and Heterogeneous Deep Neural Networks
Mohammad Qararyah, F. (2024) · Chalmers Research (Chalmers University of Technology)
Distribution of Low Latency Machine Learning \nAlgorithm
Tobias Carneiro, T. (2018) · Tampere University of Technology Institutional Repository
Efficient Deep Neural Network Computation on Processors
Yu, J. (2019) · Deep Blue (University of Michigan)
Dupuis, T. (2023) · PolyPublie (École Polytechnique de Montréal)
Compressed convolutional neural networks for sewer inspection
Determan, B. (. (2024) · Research Repository (Delft University of Technology)
Optimizing AI Accelerators through Analytical Modeling and Intelligent Mapping
Symons, A. (2025) · Lirias (KU Leuven)
Hardware Modellering en Exploratie Voor Edge AI-versnelling
Houshmand, P. (2025) · Lirias (KU Leuven)
de FRANÇA, A. L. B.-V. (2023) · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas)
PIM-AD: A PIM Architecture for CNN-Based Anomaly Detection
Lin, F., Ma, D., Xiong, Z., Tang, C., & Zhao, Z. (2025) · International Conference on Machine Learning and Computer Application (ICMLCA) · doi:10.1109/icmlca66850.2025.11336794
“Thanks to the development of emerging devices, the Processing in Memory (PIM) architecture has become one of the effective solutions to solve the memory wall challenge by integrating memory and computing [7-12].”
Benchmarking, Profiling and White-box Performance Modeling for DNN Training
Zhu, H. (2022) · TSpace (University of Toronto)
Tackling Resource Utilization In Deep Neural Network Accelerators
Uwizeyimana, I. (2022) · TSpace (University of Toronto)
A Tuning Strategy for Layer-pipeline Mapping on Autonomous Driving Accelerators
Jiang, Y., Wang, L., Tang, C., Qiao, Y., Zhou, Y., & Guan, P. (2025) · International Conference on Mechatronic Engineering and Artificial Intelligence (MEAI) · doi:10.1109/meai68126.2025.11406680
A Mapping Framework for Many-core Accelerators for Autonomous Driving
Wang, L., Jiang, Y., Tang, C., Qiao, Y., Zhou, Y., & Guan, P. (2025) · International Conference on Mechatronic Engineering and Artificial Intelligence (MEAI) · doi:10.1109/meai68126.2025.11406655
FPGA-Based Implementation of LeNet Architecture in Verilog for Disease Prediction
Dhiravidachelvi, E., Muthuvel, A., Banu, K. S., Peeroli, H., & Fathima, M. T. (2025) · IEEE International Conference on Innovations in Engineering and Next-Generation Technologies for Sustainability (ICINVENTS) · doi:10.1109/icinvents64613.2025.11401724
“Alwani et al. [5] presented techniques to generate fixed-point weights and biases compatible with FPGA memory initialization files (.mem).”
SFD: Towards Segment Fusion Dataflow for Spatial Accelerators
Wang, F., Shen, M., Ding, Y., Xiao, N., & Lu, Y. (2026) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca68181.2026.11408456
“We compare SFD against five state-of-art approaches, including CoSA [18], Fused-Layer [27], FLAT [17], FAST Fusion [20], and TileFlow [16].”
FPGA-based image processing algorithms implementations for advanced driver assistance system (本文)
ウェイ, カ., & Wei, K. (2022) · Institutional Repositories DataBase (IRDB)
Boosting Scalability and Performance: Macro Placement for Flexible 3D-Stacked ML Accelerators
Hu, J., Zhang, C., Vanna-Iampikul, P., Krishna, T., & Lim, S. K. (2026) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac66049.2026.11420699
Design of a Universal CNN Accelerator with End-to-End Deployment SupportPreprint
Chen, Z., Liu, J., Gao, J., Hong, Z., Zhang, Y., Guan, L., Yu, Q., Wang, Y., & Ma, B. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6411288
MONET: Modeling and Optimization of neural NEtwork Training from Edge to Data CentersPreprint
Morlier, J., Geens, R., Cuyckens, S., Symons, A., Verhelst, M., Gripon, V., & Léonardon, M. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2603.15002
“…aspects of scheduling: parallelism, encompassing both spatial and temporal dimensions, which leverages the intrinsic concurrency of hardware architectures, and layer-fused scheduling [27], which uses cross-layer data dependencies to minimize memory transfers and reduce the runtime of these layers.”
Morlier, J., Léonardon, M., & Gripon, V. (2026) · HAL
Efficient Implementations of Neural Network Powered by Spike Coding and Scalable Bisection Spanning
Wu, M. (2022) · Institutional Repositories DataBase (IRDB) · doi:10.34413/dr.01845
A Mapping Strategy Optimization Framework for Systolic Array Accelerators
Yuan, X., Wang, C., Wu, Q., Yue, H., Xu, H., & Wei, X. (2026) · Lecture notes in computer science · doi:10.1007/978-981-95-8399-7_24
Deep Learning Process Integration on Heterogeneous GPU/FPGA Embedded Platforms
Carballo-Hernández, W. (2022) · theses.fr (ABES) · doi:10.70675/0c5e0a51zab87z45c8z9b00z285bdd29d606
Software protections for artificial neural networks
Guiga, L. (2022) · theses.fr (ABES) · doi:10.70675/84e41563z69d5z4979z908fz31a9ea700e9a
Yoo, B., Park, S., Lee, S., & Jeon, D. (2026) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2026.3690901
“However, this approach introduces new bottlenecks associated with tile-boundary handling and interlayer data transfers [9], [10].”
UNIFY: A Reconfigurable Processing Unit for Orthogonal Dataflows in CNN Accelerators
Yang, L., Yan, Y., & Zhang, X. (2026) · International Conference on Intelligent Computing and Signal Processing (ICSP) · doi:10.1109/icsp69961.2026.11540935
“Another strategy involves fusing the execution of entire layers to reduce off-chip memory access [6], [19].”
Gohel, D., Gundrapally, A., & Choi, K. K. (2026) · Electronics · doi:10.3390/electronics15112492
FaStrA: Fast and Efficient Network Structure-Aware Layer Fusion on Deep Learning Accelerators
Choi, J., Joo, W., Gafiyatullin, A., Karpukhin, S., Kwon, K., & Kim, M. (2026) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3820047
“A widely adopted approach for reducing memory access is layer fusion [1, 4, 63], a technique that groups multiple neural network layers, thereby allowing the intermediate data between grouped layers to be forwarded through on-chip scratchpad memory rather than accessing off-chip memory.”
InfluentialWu, G., Chang, L., Wang, J., Zeng, J., Hu, Z., Zhao, X., Zhou, L., Tu, F., & Zhou, J. (2026) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas66217.2026.11562345
“Lee et al. proposed a tile-based architecture that processes the entire SR image in smaller batches to alleviate this issue [7], while Awani et al. proposed a layer-fusion strategy to explore the possibility of avoiding convolution operations to reduce memory access [8].”
Tursunaliev, U., & Tulkinov, B. (2026) · Innovation Science and Technology · doi:10.5281/zenodo.20781146
Efficient Parallel Compilation for Neural Networks on GPUs
Wang, S. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-3492-9_8
Operator Fusion for LLM Inference on the Tensix Architecture
Wu, Q., Li, K., Wang, W., Yu, J., Zhang, R. X., & Liu, L. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-3426-4_2
A Scalable and Programmable Hierarchical DNN Accelerator for Edge Inference on FPGAs
Peetakul, J., Phosri, P., Suksiri, P., & Suntiphap, P. (2026) · International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON) · doi:10.1109/ecti-con68836.2026.11608835
“As a result, conventional design approach deliver low sustained performance and energy efficiency for diverse or lightweight CNN workloads, making them less suitable for resource-constrained edge AI devices [8].”
Uçta Swin Tabanlı Dönüştürücüler için donanım hızlandırması
Esergun, Y. (2024) · Bilkent University Institutional Repository (Bilkent University)
FSGen: Agile Fused and Sparse Accelerator Generator with Accurate Power Model for LLM Applications
Jay Zhe-an, M. O. K., Zhang, Q., & Xie, Z. (2026) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1145/3770743.3804052
“Fused operation allows intermediate tensors to be passed directly to the next PE array for processing, thereby reducing costly memory accesses [1, 15].”
Prasath, M. G., & Ravi, S. (2026) · International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE) · doi:10.1109/icdcece68961.2026.11645880
“While hardware accelerators from the DianNao family [7], [8], spatial architectures like Eyeriss [9], and fused-layer accelerators [10] mitigate off-chip memory traffic via localized data reuse and scratchpad routing, they fundamentally remain value-agnostic, computing all mathematical operations…”
Fb-FaceBoxes: FPGA-based face detection using repetitive kernels and memory-access optimization
Bonakdarpour, S., Dehnavi, M., Jalali, A., & Moaiyeri, M. H. (2026) · Journal of Real-Time Image Processing · doi:10.1007/s11554-026-01954-5
“Although floating-point arithmetic provideshigheraccuracy,itdemandssignificantlymorecom-putational resources [28].”
A pipeline methodology for neural network accelerators under a throughput constraint
Oudrhiri, A., & Munier-Kordon, A. (2026) · Computers & Operations Research · doi:10.1016/j.cor.2026.107649
Quantization-based Optimization of CNN Inference on a multi-FPGA system
Pan, H. (2021) · Institutional Repositories DataBase (IRDB)
Fengshui: Demystifying Chiplet Ecosystem and Bespoke Neural Network Accelerator CodesignPreprint
Jin, H., Yang, J., Zhang, Z., Shin, J., Lyu, B., Zhang, K., Liu, Y., & Bleier, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.10970
“Known-good fusion patterns seed it—e.g., the early-layer fusion Al-wani et al. [62] show saves substantial energy in convolutional neural networks (CNNs)—and a roofline estimate seeds each boundary with the cheapest DRAM that sustains its bandwidth.”
Busoye, O. M., Busoye, T. M., & Eigbe, E.-A. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.00586
“Weight quantization [11, 16, 24], pruning [10, 13, 23], architecture search [4, 5], bit-packing [3], and fused-layer accelerators [2] all reduce total model footprint or cross-layer traffic without altering the intra-layer activation peak, and are complementary throughout.”
Communication-centric cross-stack acceleration for distributed machine learning
Li, Y. (2022) · Illinois Digital Environment for Access to Learning and Scholarship (University of Illinois at Urbana-Champaign)
CHASE: A CPU-CIM heterogeneous synergistic optimization framework for LLM workloads
Li, Y., Yang, R., & Wen, M. (2026) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2026.104044
Design-Space Exploration for Runtime-Reconfigurable FPGA-Based DNN Accelerators
Demirdag, Z., Bauer, L., Balaskas, K., Nassar, H., & Henkel, J. (2026) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3859878
Fomin, I., Koryabkin, I., & Sheludko, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2610.23328
“These models are deliberately designed so that the early high-resolution layer dominates, and further savings for them require splitting that layer’s tensors into tiles, as patch-based inference does (Lin et al., 2021; Alwani et al., 2016).”
InfluentialCNN-Accelerated Fixed-Point Inference for CIFAR-10 in SystemVerilog with Vivado Simulation
Srimanikrishna, A., & Kumar, J. R. (2026) · IEEE International Conference on Recent Advances in Electronics, Communication, Computing, Automation and Power (RECCAP) · doi:10.1109/reccap69368.2026.11724313
TranSAGE: Automating Spatial Accelerator Generation for Large-Scale Transformer Models
Zhang, C., Sun, Y., Zhang, W., Chen, Q., & Zhao, J. (2026) · International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) · doi:10.1145/3845814.3850039
Gezamenlijk ontwerp van Compilers en Hardware voor Versnelling van Deep Learning
Delm, J. V. (2026) · Lirias
Metadata
- Authors
- Alwani, M., Chen, H., Ferdman, M., & Milder, P.
- Affiliations
- Stony Brook University (US)
- Venue
- 2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO)
- Publisher
- IEEE
- Published
- 2016-10
- Pages
- 1-12
- ISSN
- 1072-4451 (ISSN-L)
- OpenAlex
- W4249932213
- Cited by
- 662
- References
- 20
- Citations by year
- 2026: 20 · 2025: 51 · 2024: 72 · 2023: 67 · 2022: 87 · 2021: 103 · 2020: 72 · 2019: 90 · 2018: 76 · 2017: 23 · 2016: 1
- FWCI
- 21.98
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE/ACM International Symposium on Microarchitecture (MICRO) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Hardware and Architecture › Embedded Systems Design Techniques
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Keywords
- CNN accelerator, layer fusion, data flow design, FPGA implementation, convolutional neural network inference
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.