A 0.32–128 TOPS, Scalable Multi-Chip-Module-Based Deep Neural Network Inference Accelerator With Ground-Referenced Signaling in 16 nm
The DOI matches its Crossref record; 26 of the 29 references it cites were found in a registry or an index.
Abstract
Custom accelerators improve the energy efficiency, area efficiency, and performance of deep neural network (DNN) inference. This article presents a scalable DNN accelerator consisting of 36 chips connected in a mesh network on a multi-chip-module (MCM) using ground-referenced signaling (GRS). While previous accelerators fabricated on a single monolithic chip are optimal for specific network sizes, the proposed architecture enables flexible scaling for efficient inference on a wide range of DNNs, from mobile to data center domains. Communication energy is minimized with large on-chip distributed weight storage and a hierarchical network-on-chip and network-on-package, and inference energy is minimized through extensive data reuse. The 16-nm prototype achieves 1.29-TOPS/mm2 area efficiency, 0.11 pJ/op (9.5 TOPS/W) energy efficiency, 4.01-TOPS peak performance for a one-chip system, and 127.8 peak TOPS and 1903 images/s ResNet-50 batch-1 inference for a 36-chip system.
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 [59]
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
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Rethinking the Inception Architecture for Computer Vision
Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., & Wojna, Z. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.308 · Cited by 32,138
Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
Chen, L.-C., Zhu, Y., Papandreou, G., Schroff, F., & Adam, H. (2018) · arXiv · doi:10.48550/arxiv.1802.02611 · Cited by 20,063
Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation
Chen, L.-C., Zhu, Y., Papandreou, G., Schroff, F., & Adam, H. (2018) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-01234-2_49 · Cited by 20,054
In-Datacenter Performance Analysis of a Tensor Processing Unit
Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., Bates, S., Bhatia, S., Boden, N., Borchers, A., et al. (2017) · Proceedings of the 44th Annual International Symposium on Computer Architecture · doi:10.1145/3079856.3080246 · Cited by 4,537 · cited 1×
“Duetothedeterministic structure of DNNs, fixed-function accelerators have the potential to further improve area effi-ciency, energy efficiency, and performance relative to CPUs and GPUs [6]–[12].”
BackgroundEyeriss: An Energy-Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks
Chen, Y.-H., Krishna, T., Emer, J. S., & Sze, V. (2017) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2016.2616357 · Cited by 3,247
Self-citationEnd to End Learning for Self-Driving Cars
Bojarski, M., Del Testa, D., Dworakowski, D., Firner, B., Flepp, B., Goyal, P., Jackel, L. D., Monfort, M., Muller, U., Zhang, J., et al. (2016) · arXiv · doi:10.48550/arxiv.1604.07316 · Cited by 3,198
EIE: efficient inference engine on compressed deep neural network
Han, S., Liu, X., Mao, H., Pu, J., Pedram, A., Horowitz, M. A., & Dally, W. J. (2016) · ACM SIGARCH Computer Architecture News · doi:10.1145/3007787.3001163 · Cited by 1,937
Self-citationMoons, B., Uytterhoeven, R., Dehaene, W., & Verhelst, M. (2017) · 2017 IEEE International Solid-State Circuits Conference (ISSCC) · doi:10.1109/isscc.2017.7870353 · Cited by 479
Research works printed but not linked: 3
- saban, 2012, Xilinx Stacked Silicon Interconnect Technology Delivers Breakthrough FPGA Capacity Bandwidth and Power Efficiency
- 2017, AMBA AXI and ACE Protocol Specification AXI3 AXI4 AXI5 ACE and ACE5
- 2019, MATCH
Cited by
A compute-in-memory chip based on resistive random-access memory
Wan, W., Kubendran, R., Schaefer, C., Eryilmaz, S. B., Zhang, W., Wu, D., Deiss, S., Raina, P., Qian, H., Gao, B., et al. (2022) · Nature · doi:10.1038/s41586-022-04992-8 · Cited by 943
Self-citationMemristive technologies for data storage, computation, encryption, and radio-frequency communication
Lanza, M., Sebastian, A., Lu, W. D., Le Gallo, M., Chang, M.-F., Akinwande, D., Puglisi, F. M., Alshareef, H. N., Liu, M., & Roldan, J. B. (2022) · Science · doi:10.1126/science.abj9979 · Cited by 708
“Reference (57) (133) (41) (59) (86) (134) (135)”
Mixed-Signal Computing for Deep Neural Network Inference
Murmann, B. (2021) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2020.3020286 · Cited by 160
“Beyond B = 8, the estimate exceeds 200 fJ/MAC, which is also reachable with purely digital implementations [27].”
BackgroundMethodInfluentialA Survey on Deep Learning Hardware Accelerators for Heterogeneous HPC Platforms
Silvano, C., Ielmini, D., Ferrandi, F., Fiorin, L., Curzel, S., Benini, L., Conti, F., Garofalo, A., Zambelli, C., Calore, E., et al. (2025) · ACM Computing Surveys · doi:10.1145/3729215 · Cited by 262
“Finally, as a representative chiplet-based DNN hardware accelerator, we report Simba [182, 227].”
PIMCA: A Programmable In-Memory Computing Accelerator for Energy-Efficient DNN Inference
Zhang, B., Yin, S., Kim, M., Saikia, J., Kwon, S., Myung, S., Kim, H., Kim, S. J., Seo, J.-S., & Seok, M. (2022) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2022.3211290 · Cited by 80
“To tackle these challenges, recently, a number of digital DNN hardware accelerators have been proposed [5], [6], [7], [8].”
BackgroundKhoramnejad, F., & Hossain, E. (2024) · IEEE Communications Surveys and Tutorials · doi:10.1109/comst.2025.3535554 · Cited by 91
“As discussed in [102], future paradigms in DL accelerator design envision the development of accelerators for sparse matrices [103], 3D-stacked processing-in-memory [104], in-memory computing [105], neuromorphic accelerators [106], and multi-chip modules [107].”
NN-Baton: DNN Workload Orchestration and Chiplet Granularity Exploration for Multichip Accelerators
Tan, Z., Cai, H., Dong, R., & Ma, K. (2021) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1109/isca52012.2021.00083 · Cited by 104
“Simba [55, 62, 78, 79] is a pioneering work to implement a DNN accelerator using chiplet technology.”
BackgroundMethodSurvey on chiplets: interface, interconnect and integration methodology
Ma, X., Wang, Y., Wang, Y., Cai, X., & Han, Y. (2022) · CCF Transactions on High Performance Computing · doi:10.1007/s42514-022-00093-0 · Cited by 71
Gemini: Mapping and Architecture Co-exploration for Large-scale DNN Chiplet Accelerators
Cai, J., Wu, Z., Peng, S., Wei, Y., Tan, Z., Shi, G., Gao, M., & Ma, K. (2024) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca57654.2024.00022 · Cited by 103
Inter-layer Scheduling Space Definition and Exploration for Tiled Accelerators
Cai, J., Wei, Y., Wu, Z., Peng, S., & Ma, K. (2023) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1145/3579371.3589048 · Cited by 103
“While most existing tiled accelerators have supported the inter-layer scheduling behaviors mentioned above efficiently [7, 25, 50, 52, 56, 64], their architecture details, such as control logic, NoC architecture, etc.”
BackgroundPrabhu, K., Gural, A., Khan, Z. F., Radway, R. M., Giordano, M., Koul, K., Doshi, R., Kustin, J. W., Liu, T., Lopes, G. B., et al. (2022) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2022.3140753 · Cited by 79
Self-citationDigital Versus Analog Artificial Intelligence Accelerators: Advances, trends, and emerging designs
Seo, J.-S., Saikia, J., Meng, J., He, W., Suh, H.-S., Anupreetham, A., Liao, Y., Hasssan, A., & Yeo, I. (2022) · IEEE Solid-State Circuits Magazine · doi:10.1109/mssc.2022.3182935 · Cited by 77
“Analog/Mixed-Signal AI Accelerators From the digital AI accelerators in the literature that reported power/energy breakdown [27], [28] ( Figure 3), it can be seen that data access energy to/ from on-chip SRAMs constitute two thirds or more of the total system power/energy.”
BackgroundA 95.6-TOPS/W Deep Learning Inference Accelerator With Per-Vector Scaled 4-bit Quantization in 5 nm
Keller, B., Venkatesan, R., Dai, S. X., Tell, S. G., Zimmer, B., Sakr, C., Dally, W. J., Gray, C. T., & Khailany, B. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3234893 · Cited by 74
Self-citationFilament Engineering of Two‐Dimensional h‐BN for a Self‐Power Mechano‐Nociceptor System
Ding, G., Chen, R., Xie, P., Yang, B., Shang, G., Liu, Y., Gao, L., Mo, W., Zhou, K., Han, S., et al. (2022) · Small · doi:10.1002/smll.202200185 · Cited by 67
Liu, J., Fan, J., Zhong, Z., Qiu, H., Xiao, J., Zhou, Y., Zhu, Z., Dai, G., Wang, N., Liu, Q., et al. (2023) · IEEE Transactions on Biomedical Circuits and Systems · doi:10.1109/tbcas.2023.3276782 · Cited by 73
“To address this issue, AI processors have been developed to accelerate the computation of neural network for general AI applications [19], [20], [21], [22].”
MethodXuan, L., Un, K.-F., Lam, C., & Martins, R. P. (2022) · IEEE Transactions on Circuits & Systems II Express Briefs · doi:10.1109/tcsii.2022.3180553 · Cited by 66
“Recently, a plethora of works contributed to the design of energy-efficient DNN accelerators for graphics processing units (GPUs) [2], application-specific integrated circuits (ASICs) [3-7], and fieldprogrammable gate arrays (FPGAs) [8-10].”
BackgroundRecent Progress in Neuromorphic Computing from Memristive Devices to Neuromorphic Chips
Xiao, Y., Gao, C., Jin, J., Sun, W., Wang, B., Bao, Y., Liu, C., Huang, W., Zeng, H., & Yu, Y. (2024) · Advanced Devices & Instrumentation · doi:10.34133/adi.0044 · Cited by 68
The Big Chip: Challenge, model and architecture
Han, Y., Xu, H., Lu, M., Wang, H., Huang, J., Wang, Y., Wang, Y., Feng, M., Liu, Q., Liu, M., et al. (2023) · Fundamental Research · doi:10.1016/j.fmre.2023.10.020 · Cited by 62
A Scalable Methodology for Designing Efficient Interconnection Network of Chiplets
Feng, Y., Xiang, D., & Ma, K. (2023) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca56546.2023.10070981 · Cited by 41
“present the multi-chipmodule-based DNN inference accelerator [4] [6], which can scale from 0.”
BackgroundTambe, T., Yang, E.-Y., Ko, G. G., Chai, Y., Hooper, C. R. C., Donato, M., Whatmough, P. N., Rush, A. M., Brooks, D. M., & Wei, G.-Y. (2022) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2022.3179303 · Cited by 46
A Survey of Computing-in-Memory Processor: From Circuit to Application
Sun, W., Yue, J., He, Y., Huang, Z., Wang, J., Jia, W., Li, Y., Lei, L., Jia, H., & Liu, Y. (2023) · IEEE Open Journal of the Solid-State Circuits Society · doi:10.1109/ojsscs.2023.3328290 · Cited by 49
“Digital spatial architectures also pay costs for data readout, as an instance, the accumulation buffers for PEs in [53].”
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
“DNN decoupled accelerators represent a well-studied topic [5], [17], [41], [71], [80].”
BackgroundNear-Sensor Distributed DNN Processing for Augmented and Virtual Reality
Pinkham, R., Berkovich, A., & Zhang, Z. (2021) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2021.3121259 · Cited by 31
“However, with advanced chiplet-based [37]–[39] and”
Fair and Comprehensive Benchmarking of Machine Learning Processing Chips
Burr, G. W., Lim, S., Murmann, B., Venkatesan, R., & Verhelst, M. (2021) · IEEE Design and Test · doi:10.1109/mdat.2021.3063366 · Cited by 32
Self-citationSWAP: A Server-Scale Communication-Aware Chiplet-Based Manycore PIM Accelerator
Sharma, H., Mandal, S. K., Doppa, J. R., Ogras, U. Y., & Pande, P. P. (2022) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2022.3197500 · Cited by 29
“For a server-scale scenario, the number of DNN parameters is in the order of hundreds of millions [20].”
BackgroundJia, T., Mantovani, P., dos Santos, M. C., Giri, D., Zuckerman, J. D., Loscalzo, E. J., Cochet, M., Swaminathan, K., Tombesi, G., Zhang, J. J., et al. (2022) · European Solid-State Circuits Conference (ESSCIRC) · doi:10.1109/esscirc55480.2022.9911456 · Cited by 26
“11 compares this work to prior tile-based chip designs, which feature homogeneous arrays of either processors [11-12] or accelerators [13].”
MethodSMIV: A 16-nm 25-mm² SoC for IoT With Arm Cortex-A53, eFPGA, and Coherent Accelerators
Lee, S. K., Whatmough, P. N., Donato, M., Ko, G. G., Brooks, D. M., & Wei, G.-Y. (2021) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2021.3115466 · Cited by 30
IECA: An In-Execution Configuration CNN Accelerator With 30.55 GOPS/mm² Area Efficiency
Huang, B., Huan, Y., Chu, H., Xu, J., Liu, L., Zheng, L., & Zou, Z. (2021) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2021.3108762 · Cited by 27
“Recent designs of CNN accelerators [6]–[13] mainly focus on mining the parallelism of processing on reusable data batches, in time or space.”
BackgroundResultReview of chiplet-based design: system architecture and interconnection
Liu, Y., Li, X., & Yin, S. (2024) · Science China Information Sciences · doi:10.1007/s11432-023-3926-8 · Cited by 28
Enhancing Collective Communication in MCM Accelerators for Deep Learning Training
Laskar, S., Majhi, P., Kim, S., Mahmud, F., Muzahid, A., & Kim, E. J. (2024) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca57654.2024.00069 · Cited by 32
An Eight-Core 1.44-GHz RISC-V Vector Processor in 16-nm FinFET
Schmidt, C., Wright, J., Wang, Z., Chang, E. Y., Ou, A. J., Bae, W., Huang, S., Milovanović, V., Flynn, A. M., Richards, B. C., et al. (2021) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2021.3118046 · Cited by 21
Sharma, H., Pfromm, L., Topaloglu, R. O., Doppa, J. R., Ogras, U. Y., Kalyanraman, A., & Pande, P. P. (2023) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3608098 · Cited by 24
“Most chip vendors and foundries are moving towards non-monolithic alternatives such as 2.5D interposer-based systems to partition the on-chip resources into smaller discrete cores called chiplets [ 1 , 13 , 14 ].”
BackgroundFeng, K., Kong, T., Koul, K., Melchert, J., Carsello, A., Liu, Q., Nyengele, G., Strange, M., Zhang, K., Nayak, A., et al. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3313116 · Cited by 29
Self-citationHeterogeneous Die-to-Die Interfaces: Enabling More Flexible Chiplet Interconnection Systems
Feng, Y., Xiang, D., & Ma, K. (2023) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1145/3613424.3614310 · Cited by 19
Modeling and Optimization of SRAM-based In-Memory Computing Hardware Design
Saikia, J., Yin, S., Cherupally, S. K., Zhang, B., Meng, J., Seok, M., & Seo, J.-S. (2021) · Design, Automation, and Test in Europe · doi:10.23919/date51398.2021.9473973 · Cited by 21
“However, the energy/power breakdown reported in [5]– [7] shows that memory access and data movement accounts for a dominant portion of the total on-chip energy/power consumption.”
Tang, W., Cho, S.-G., Hoang, T. T., Botimer, J., Zhu, W. Q., Chang, C.-C., Lu, C., Zhu, J., Tao, Y., Wei, T., et al. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3343457 · Cited by 17
“In [6], 36 DNN accelerator chiplets are integrated on an organic substrate using a 25-Gb/s/pin ground-referenced signaling (GRS) interface [7].”
BackgroundMethodSumbul, H. E., Seo, J.-S., Morris, D. H., & Beigné, É. (2023) · IEEE Micro · doi:10.1109/mm.2023.3338059 · Cited by 16
Kosuge, A., Hamada, M., & Kuroda, T. (2021) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2021.3114179 · Cited by 19
Huang, Y., Tan, Y., Kang, Y., Chen, Y., Tang, Y., & Jiang, T. (2024) · Science China Information Sciences · doi:10.1007/s11432-023-3888-0 · Cited by 18
“Detailed performance parameter comparisons between the RRAM-based CIM systems and CMOS AI inference accelerators are listed in Table 1 [92–96].”
MethodA Hybrid-Mode On-Chip Router for the Large-Scale FPGA-Based Neuromorphic Platform
Ding, C., Huan, Y., Jia, H., Yan, Y., Yang, F., Liu, L., Shen, M., Zou, Z., & Zheng, L. (2022) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2022.3145016 · Cited by 13
CASSANN-v2: A high-performance CNN accelerator architecture with on-chip memory self-adaptive tuning
Liu, F., Qiao, R., Chen, G., Gong, G., & Lu, H. (2022) · IEICE Electronics Express · doi:10.1587/elex.19.20220124 · Cited by 13
Advancements in Artificial Intelligence Circuits and Systems (AICAS)
Miller, T. I., Durlik, I., Kostecka, E., Mitan-Zalewska, P., Sokołowska, S., Cembrowska-Lech, D., & Łobodzińska, A. (2023) · Electronics · doi:10.3390/electronics13010102 · Cited by 16
Jiao, B., Xu, L., Yu, X., Hai-tao, Y., Zhu, H., Wang, Y., Zhu, J., Wen, D., Wang, L., Tao, J., et al. (2024) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2024.3419579 · Cited by 12
“As a result, it is power-hungry for the on-package network to convert analog signals back to digital-first, perform message routing, and convert again for bypass messages [30].”
BackgroundA Heterogeneous Chiplet Architecture for Accelerating End-to-End Transformer Models
Sharma, H., Dhingra, P., Doppa, J. R., Ogras, U. Y., & Pande, P. P. (2025) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3718487 · Cited by 22
“𝑞 &;6 = w1, if chiplet i,j communicate across link k.”
Cherupally, S. K., Meng, J., Rakin, A. S., Yin, S., Yeo, I., Yu, S., Fan, D., & Seo, J.-S. (2021) · Semiconductor Science and Technology · doi:10.1088/1361-6641/ac461f · Cited by 12
“On the hardware side, many digital accelerators have been proposed that address these issues by using optimized dataflows, systolic arrays of multiply-accumulate (MAC) units, and memory hierarchies [4–6].”
Jiao, B., Zhu, H., Zeng, Y., Li, Y., Liao, J., Jia, S., Tian, M., Chen, Z., Zhu, J., Wen, D., et al. (2025) · Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference · doi:10.1109/isscc49661.2025.10904819 · Cited by 11
“To 25 $ . validate both approaches, an NPU chiplet, equipped with a RISC-V core [7] and a previous compute-in-memory accelerator [8], was developed to support both interfaces.”
Liu, J., Xie, Z., Liu, X., Wang, X., Xiao, J., Guo, C., Fan, J., Liu, Q., Zhu, Z., Li, S., et al. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3555512 · Cited by 11
“With general neural network (NN) computing platforms such as GPUs and AI processors, the power consumption can easily go up to hundreds of milliwatts or several watts, posing significant challenges to meeting the ultra-low power consumption requirement of wearable devices [12], [13], [14].”
Characterization and Mitigation of Relaxation Effects on Multi-level RRAM based In-Memory Computing
He, W., Shim, W., Yin, S., Sun, X., Fan, D., Yu, S., & Seo, J.-S. (2021) · IEEE International Reliability Physics Symposium proceedings · doi:10.1109/irps46558.2021.9405228 · Cited by 15
“…DNN computations are dominated by multiply-and-accumulate (MAC) operations, but the overall energy consumption of DNN inference hardware has been dominated by memory access and data communication [2], due to the separation of conventional memories with row-by-row access and dedicated MAC engines.”
BackgroundImproving DNN Hardware Accuracy by In-Memory Computing Noise Injection
Cherupally, S. K., Meng, J., Rakin, A. S., Yin, S., Seok, M., Fan, D., & Seo, J.-S. (2021) · IEEE Design and Test · doi:10.1109/mdat.2021.3139047 · Cited by 13
HPPI: A High-Performance Photonic Interconnect Design for Chiplet-Based DNN Accelerators
Li, G., & Ye, Y. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3328828 · Cited by 10
“Chiplet-based accelerator designs using MCM have been proposed to implement high-performance DNN inference [8], [9], [10], [11], [12].”
MethodWang, Y., Wu, Z., Wu, W., Liu, L., Hu, Y., Wei, S., Tu, F., & Yin, S. (2024) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2024.3406569 · Cited by 11
M2M: A Fine-Grained Mapping Framework to Accelerate Multiple DNNs on a Multi-Chiplet Architecture
Zhang, J., Wang, X., Ye, Y., Lyu, D., Xiong, G., Xu, N., Lian, Y., & He, G. (2024) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2024.3438549 · Cited by 9
“The increasing number of chiplets leads to problems such as distributed memory access, nonuniform latency, and imbalanced traffic patterns [21], [22].”
BackgroundMethodInfluentialNimbekar, A., Vatti, C. S., Dinesh, Y. V. S., Singh, S., Gupta, T., Chandrapu, R. R., & Acharyya, A. (2022) · IEEE International System-on-Chip Conference (SOCC) · doi:10.1109/socc56010.2022.9908073 · Cited by 9
“[ 8], [10], [14] [13] does the computation for the input data samples with zero values, which results in more power consumption.”
BackgroundInfluentialWang, Z., Nalla, P. S., Sun, J., Goksoy, A. A., Mandal, S. K., Seo, J.-S., Chhabria, V. A., Zhang, J. J., Chakrabarti, C., Ogras, U. Y., et al. (2025) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2025.3531348 · Cited by 11
“Symbols are from [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25]. computing, the mapping and optimizing target AI algorithms on an HI system present intriguingly complex challenges, attributed to the abundance of design variables involved in this…”
Srinivasa, S., Kurian, D. J., Aseron, P. A., Budhkar, P., Radhakrishnan, A., Lopez, A. C., Sundaram, J., Honkote, V., Azarenkov, L., Lake, D., et al. (2025) · Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference · doi:10.1109/isscc49661.2025.10904798 · Cited by 8
Zhang, C., You, Y., Wang, N., Park, J., & Zhang, L. (2025) · IEEE Journal on Emerging and Selected Topics in Circuits and Systems · doi:10.1109/jetcas.2025.3575272 · Cited by 14
Learning to Design Accurate Deep Learning Accelerators with Inaccurate Multipliers
Jain, P., Huda, S., Maas, M., Gonzalez, J. E., Stoical, I., & Mirhoseini, A. (2022) · Design, Automation, and Test in Europe · doi:10.23919/date54114.2022.9774607 · Cited by 14
“[22] report that over 30% of PE energy is consumed by arithmetic units [17].”
BackgroundAchieving Datacenter-scale Performance through Chiplet-based Manycore Architectures
Sharma, H., Mandal, S. K., Doppa, J. R., Ogras, U. Y., & Pande, P. P. (2023) · Design, Automation, and Test in Europe · doi:10.23919/date56975.2023.10137125 · Cited by 8
Hybrid Stochastic Number and Its Neural Network Computation
Li, H., & Chen, Y. (2023) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2023.3332170 · Cited by 10
BlitzCoin: Fully Decentralized Hardware Power Management for Accelerator-Rich SoCs
Cochet, M., Swaminathan, K., Loscalzo, E. J., Zuckerman, J. D., dos Santos, M. C., Giri, D., Buyuktosunoglu, A., Jia, T., Brooks, D. M., Wei, G.-Y., et al. (2024) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1109/isca59077.2024.00063 · Cited by 7
“According to prior state-of-the-art tile-based ASICs [47], [63], [64], as well as our own estimations, the NoC accounts for roughly 5-10% of the power of such SoCs.”
BackgroundAxLaM: energy-efficient accelerator design for language models for edge computing
Glint, T., Mittal, B., Sharma, S., Ronak, A. Q., Goud, A., Kasture, N., Momin, Z., Krishna, A., & Mekie, J. (2025) · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · doi:10.1098/rsta.2023.0395 · Cited by 7
“Simba [8] introduced a scalable multi-chip framework optimized for convolutional neural networks (CNNs), but it faces challenges with transformer models owing to differing computational patterns.”
Martins, R. P., Mak, P., Sin, S., Law, M. K., Zhu, Y., Lü, Y., Yin, J., Chan, C., Chen, Y., Un, K.-F., et al. (2021) · Foundations and Trends® in Integrated Circuits and Systems · doi:10.1561/3500000007 · Cited by 8
A NoC-Based Spatial DNN Inference Accelerator With Memory-Friendly Dataflow
Zhu, L., Fan, W., Dai, C., Zhou, S., Xue, Y., Lu, Z., Li, L., & Fu, Y. (2023) · IEEE Design and Test · doi:10.1109/mdat.2023.3310199 · Cited by 7
Lee, S. M., Kim, H., Yeon, J., Lee, J., Choi, Y., Kim, M., Park, C.-J., Jang, K., Kim, Y., Kim, Y., et al. (2022) · IEEE Open Journal of the Solid-State Circuits Society · doi:10.1109/ojsscs.2022.3216798 · Cited by 6
“The results are better or comparable to the performance of the previous work [28].”
ResultJiao, B., Zhu, H., Zhang, J., Wang, S., Kang, X., Zhang, L., Wang, M., & Chen, C. (2021) · ACM Great Lakes Symposium on VLSI (GLSVLSI) · doi:10.1145/3453688.3461499 · Cited by 6
Hardware-Software Codesign of DNN Accelerators Using Approximate Posit Multipliers
Glint, T., Prasad, K., Dagli, J., Gandhi, K., Gupta, A., Patel, V., Shah, N., & Mekie, J. (2023) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3566097.3567866 · Cited by 8
“2 Background DNN workloads consisting of convolution layers (CL) and fully connected layers (FCL) can be processed using DNN accelerators [1, 2, 16].”
BackgroundMethodInfluentialLi, J., Li, K., Un, K.-F., Yu, W.-H., Martins, R. P., & Mak, P. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3532544 · Cited by 8
“Normalized to 28 nm, this work also performs better in the area efficiency, from 1.5 × [33] to 1.8 × [20].”
InfluentialJain, V., Cavalcante, M., Bruschi, N., Rogenmoser, M., Benz, T., Kurth, A., Rossi, D., Benini, L., & Verhelst, M. (2023) · ACM/IEEE Design Automation Conference (DAC) · doi:10.1109/dac56929.2023.10247800 · Cited by 7
“SpiNNaker [16] × × × × 5 (async) Reza et al [17] × × × × 4000 MCM [18] × × × × 35 MC-NoC [6] × × × × 2368 NeuNoC [19] × × × × TETRIS [20] × × × × PUMA [21] × × × × OpenSoC [22] ✓ × × ✓ ESP-SoC [4] ✓ × × Limited 351 Celerity [23] ✓ × × Limited 80 FlexNoC [24] × × × Constellation [25] ✓ × × ✓ Andreas et al.”
Possibilities and Limitations of Memristor Crossbars for Neuromorphic Computing
Telminov, O. A., & Gornev, E. (2022) · Scientific School Dynamics of Complex Networks and their Applications (DCNA) · doi:10.1109/dcna56428.2022.9923302 · Cited by 4
“Original ways of interaction between them are proposed, e.g., Ground-Referenced Signaling [41].”
BackgroundA Dataflow-Aware Network-on-Interposer for CNN Inferencing in the Presence of Defective Chiplets
Sharma, H., Ogras, U. Y., Kalyanaraman, A., & Pande, P. P. (2024) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2024.3447210 · Cited by 3
“Given the need to execute various CNN inferencing tasks simultaneously, modern-day servers and high-end processors must be designed to execute multiple concurrent applications [29], [30].”
BackgroundPerformance Implications of Multi-Chiplet Neural Processing Units on Autonomous Driving Perception
Odema, M., Chen, L., Kwon, H., & Faruque, M. A. A. (2025) · Design, Automation, and Test in Europe · doi:10.23919/date64628.2025.10993228 · Cited by 4
“Our findings demonstrate that the combination of throughput matching algorithm and heterogeneous integration offer desirable performance trade-offs from the multi-chiplet module despite the added NoP costs, motivating further studies on adopting forthcoming chiplet-based NPU architectures in the…”
InfluentialQin, Y., Purdy, R., Probst, A., Lin, C., & Zhu, J.-G. (2023) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2023.3241631 · Cited by 4
“It is worth pointing out that the high area efficiency reported in [38] is only for a single matrix–matrix multiplication accelerator, whereas the numbers reported in [39] and [41] and our chip are measured on the overall network core.”
BackgroundInfluentialREDRAW: Fast and Efficient Hardware Accelerator with Reduced Reads And Writes for 3D UNet
Glint, T., Awasthi, M., & Mekie, J. (2023) · Design, Automation, and Test in Europe · doi:10.23919/date56975.2023.10137190 · Cited by 3
“Simba [14] is the state-of-the-art and data-agnostic Hardware Accelerator (HA) architecture for CNN-based inferencing on As shown in Fig.”
BackgroundMachine Learning Hardware Design for Efficiency, Flexibility, and Scalability [Feature]
Jie-Fang, Z., & Zhang, Z. (2023) · IEEE Circuits and Systems Magazine · doi:10.1109/mcas.2023.3302390 · Cited by 3
“22, where up to 36 DNN chiplets can be integrated in one MCP to scale up the system as needed [52].”
BackgroundMethodInfluentialReFOCUS: Reusing Light for Efficient Fourier Optics-Based Photonic Neural Network Accelerator
Li, S., Yang, H., Wong, C. W., Sorger, V. J., & Gupta, P. (2023) · IEEE/ACM International Symposium on Microarchitecture (MICRO) · doi:10.1145/3613424.3623798 · Cited by 5
“Tobetterdemonstrate the advantage of ReFOCUS, we conduct a comparison with some well-known digital accelerators from both industry and academia, namely, the NVIDIA H100 GPU [3], Google TPU V3 [1], Simba [51], and a design from JSSC 20 [70], on the relatively large ResNet-50 network.”
MethodGlint, T., Awasthi, M., & Mekie, J. (2024) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac58780.2024.10473935 · Cited by 8
“Simba [7] (CHA:SOTA) from the CHA paradigm with a peak throughput of 4.01 TOPS and implemented with 16-bit word size is also considered for comparison.”
BackgroundMethodInfluentialTime Reversal for Near-Field Communications on Multi-Chip Wireless Networks
Rodríguez-Galán, F., Bandara, A., de Santana, E. P., Bolívar, P. H., Alarcón, E., & Abadal, S. (2024) · IEEE Communications Magazine · doi:10.1109/mcom.002.2300759 · Cited by 3
“However, when scaling towards massive architectures and reaching beyond the boundaries of a simple chip, where wired links become considerably slower [2], NoC exhibits important latency and energy consumption issues.”
BackgroundAnalysis of Conventional, Near-Memory, and In-Memory DNN Accelerators
Glint, T., Jha, C. K., Awasthi, M., & Mekie, J. (2023) · IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) · doi:10.1109/ispass57527.2023.00049 · Cited by 4
“They can be broadly classified into Conventional Hardware Accelerators (CHA) [8], [9], Near Data Processors (NDP) [7], [9], [11], and Processing in Memory (PIM) Accelerators [6], [10] based on [12].”
BackgroundMethodInfluentialLei, L., Zhu, Y., Zou, X., Zhang, Z., Zhao, Z., He, Y., Du, G., Guan, Z., Yang, H., Liu, Y., et al. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3568186 · Cited by 3
Onyx: A 12-nm Programmable Accelerator for Dense and Sparse Applications
Koul, K., Hsu, O., Mei, Y., Ravipati, S. G., Strange, M., Melchert, J., Carsello, A., Kong, T., Chen, P., Ke, H., et al. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3604724 · Cited by 4
“A PPLICATIONS ranging from machine learning to scientific computing leverage hardware accelerators to improve performance and energy efficiency [1], [2].”
Self-citationVerma, P., Gupta, T., Das, P., Nali, A. R., Hunsigida, V., & Acharyya, A. (2024) · IEEE International System-on-Chip Conference (SOCC) · doi:10.1109/socc62300.2024.10737722 · Cited by 2
“While CNNs are inherently complex, numerous efforts have been made to reduce their computational complexity, particularly for mobile and edge devices, as demonstrated in works such as [4] and [5].”
BackgroundAn Efficient SI/PI Co-Optimization and Analysis Method for High-Speed Pin Arrays Design
Xu, R., Li, D., Song, R., Wang, H., Zhang, L., & Li, E. (2025) · IEEE Transactions on Components Packaging and Manufacturing Technology · doi:10.1109/tcpmt.2025.3566074 · Cited by 2
Wang, S., Yu, Q., Xie, T., Ma, C., & Pei, J. (2023) · Frontiers in Neuroscience · doi:10.3389/fnins.2023.1168864 · Cited by 2
Impact of Optimal Design Point on Performance Metrics of DNN accelerators in FPGA
Glint, T., Gupta, A., Giftson, D., Shah, G., Patel, V., Chudasama, R., More, S., & Mekie, J. (2023) · IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) · doi:10.1109/ispass57527.2023.00042 · Cited by 1
“Simba is made up of 16 Processing Elements (PE), with each PE having eight vector MACs.”
BackgroundMethodInfluentialDataflow-Aware PIM-Enabled Manycore Architecture for Deep Learning Workloads
Sharma, H., Narang, G., Doppa, J. R., Ogras, U. Y., & Pande, P. P. (2024) · Design, Automation, and Test in Europe · doi:10.23919/date58400.2024.10546730 · Cited by 1
Tang, W., Cho, S.-G., Jie-Fang, Z., & Zhang, Z. (2025) · IEEE Solid-State Circuits Magazine · doi:10.1109/mssc.2025.3549361 · Cited by 1
“Similarly, Nvidia’s deep neural net work (DNN) MCM [30], [31] integrates up to 36 DNN chiplets in an organic package.”
Srinivasa, S., Kurian, D. J., Aseron, P. A., Budhkar, P., Honkote, V., Lake, D., Moiseev, M., Muñoz, R., Srinivasan, A., & Karnik, T. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3613339 · Cited by 1
“Systems integrating multiple silicon dies (chiplets) with dimensions approaching or exceeding the reticle limit are becoming an inevitable architectural trend for achieving the necessary acceleration [1], [2], [3].”
Wan, R., Lin, C.-T., Zhang, B., Tsividis, Y., & Seok, M. (2025) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2025.3622309 · Cited by 1
“Compared with traditional digital processors [5], [6], [7], some studies leverage analog-mixed-signal (AMS) hardware to enhance computational and area e ffi ciency [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23].”
Li, S., Song, C., Kim, H.-Y., Kim, N. S., & Shanbhag, N. R. (2025) · IEEE Transactions on Circuits and Systems I Regular Papers · doi:10.1109/tcsi.2025.3617327 · Cited by 1
“Recent benchmarking e ff orts of IMC, particularly at the bank-level, have demonstrated energy e ffi ciency improvements of up to 13 × and compute density enhancements of up to 7 × compared to digital accelerators [21], [22], [23], [24], [25].”
Prabhu, K., Yu, J. J., Pan, X., & Raina, P. (2026) · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · doi:10.1098/rsta.2024.0514 · Cited by 1
Self-citationNetTLM-DSE: Design Space Exploration for DNN Layer-Pipeline Spatial Mappings
Kim, S., Heo, J., Shin, H., Lee, J., Park, S., & Park, C. S. (2026) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac66049.2026.11420537 · Cited by 1
“Recent works such as Simba [9], [18], TPUv4 [19], and Tianjic [20] have introduced large-scale accelerator network architectures to meet the increasing demand for high-throughput DNN inference.”
Marium, S. M., & Chen, S. (2026) · Microelectronics · doi:10.3390/microelectronics2020010 · Cited by 1
Fine spatial-temporal density mapping with optimized approaches for many-core system
Wang, S., Gao, Y., Seng, B., Pei, J., Zhang, Y., & Huang, J. (2025) · Frontiers in Neuroscience · doi:10.3389/fnins.2025.1512926 · Cited by 1
Self-citationChen, X., Dong, H., Mei, J., Li, J., Tian, Y., Yin, J., & Xiao, F. (2025) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas56072.2025.11043704 · Cited by 1
IPDR: An Inter-Chiplet Priority-Driven Deadlock Resolution for 2-D/2.5-D Multichiplet Systems
Zhang, J., CHEN, Z., Ye, Y., Chen, H., Han, X., Jiang, J., 绳伟光, Xu, N., Lian, Y., & He, G. (2025) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2025.3583289 · Cited by 1
“For example, in SIMBA [6], tape-out results in TSMC’s 16 nm FinFET process indicate that the NoP interconnect latency is 20 ns per hop.”
InfluentialLink Bit-Error-Rate Requirement Analysis for Deep Neural Network Accelerators
Lee, J., Kim, G., Park, J., & Bae, H. (2021) · IEEE International Symposium on Circuits and Systems proceedings · doi:10.1109/iscas51556.2021.9401112 · Cited by 1
Li, X. (2022) · Wireless Communications and Mobile Computing · doi:10.1155/2022/4363937 · Cited by 1
“In recent years, research on single-chip multicore processor architecture has gradually increased, with many students using it for some connections; for example, the new server group is a multistudent connection [1].”
BackgroundRetractedKlinefelter, A. (2020) · IEEE Solid-State Circuits Magazine · doi:10.1109/mssc.2020.3021840
Self-citationIPA: Floorplan-Aware SystemC Interconnect Performance Modeling and Generation for HLS-based SoCs
Pinckney, N., Venkatesan, R., Keller, B., & Khailany, B. (2021) · International Conference on Computer Aided Design · doi:10.1109/iccad51958.2021.9643499
“11: a centralized monolithic GB along the right edge of the floorplan area (similar to [20]) or distributed among the PEs.”
BackgroundSelf-citationYu, C., Kim, H., Kim, B., & Kim, T. T.-H. (2022) · Springer eBooks · doi:10.1007/978-3-030-98781-7_2
Metadata
- Authors
- Zimmer, B., Venkatesan, R., Shao, Y. S., Clemons, J., Fojtik, M., Jiang, N., Keller, B., Klinefelter, A., Pinckney, N., Raina, P., Tell, S. G., Zhang, Y., Dally, W. J., Emer, J. S., Gray, C. T., Keckler, S. W., & Khailany, B.
- Affiliations
- Nvidia (United States) (US)
- University of California, Berkeley (US)
- Stanford University (US)
- ORCID
- Venue
- IEEE Journal of Solid-State Circuits
- Published
- 2020-04
- Volume
- 55
- Issue
- 4
- Pages
- 920-932
- ISSN
- 0018-9200 (ISSN-L)
- 1558-173X
- OpenAlex
- W2999356938
- Semantic Scholar
- fe261bfd9d74f97268b3a9d9b700fdf5340ef434
- Cited by
- 122
- References
- 29
- Citations by year
- 2026: 15 · 2025: 25 · 2024: 16 · 2023: 28 · 2022: 20 · 2021: 15 · 2020: 3
- FWCI
- 6.42
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Journal of Solid-State Circuits (publishedVersion)
- Funders
- U.S. Government through the DARPA CRAFT Program
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Computer Networks and Communications › Interconnection Networks and Systems
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Keywords
- multichip modules, network-on-chip, data reuse, energy efficiency, area efficiency, 16nm CMOS
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.