Minerva: enabling low-power, highly-accurate deep neural network accelerators
The DOI matches its Crossref record; 69 of the 70 references it cites were found in a registry or an index, and 1 are not checked yet.
Abstract
The continued success of Deep Neural Networks (DNNs) in classification tasks has sparked a trend of accelerating their execution with specialized hardware. While published designs easily give an order of magnitude improvement over general-purpose hardware, few look beyond an initial implementation. This paper presents Minerva, a highly automated co-design approach across the algorithm, architecture, and circuit levels to optimize DNN hardware accelerators. Compared to an established fixed-point accelerator baseline, we show that fine-grained, heterogeneous datatype optimization reduces power by 1.5×; aggressive, inline predication and pruning of small activity values further reduces power by 2.0×; and active hardware fault detection coupled with domain-aware error mitigation eliminates an additional 2.7× through lowering SRAM voltages. Across five datasets, these optimizations provide a collective average of 8.1× power reduction over an accelerator baseline without compromising DNN model accuracy. Minerva enables highly accurate, ultra-low power DNN accelerators (in the range of tens of milliwatts), making it feasible to deploy DNNs in power-constrained IoT and mobile devices.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
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 · cited 1×
“These methods have gained popularity over the last ten years thanks to empirical achievements on a wide range of tasks including speech recognition [2] [3], computer vision [4], and natural language processing [5] [6].”
BackgroundLeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
(2014) · Cited by 33,419
Multilayer feedforward networks are universal approximators
Hornik, K., Stinchcombe, M., & White, H. (1989) · Neural Networks · doi:10.1016/0893-6080(89)90020-8 · Cited by 21,728 · cited 1×
“Deep Neural Networks (DNNs) are Machine Learning (ML) methods that learn complex function approximations from input/output examples [1].”
BackgroundNeural networks for pattern recognition
(1994) · Choice Reviews Online · doi:10.5860/choice.31-5500 · Cited by 18,538
Sequence to Sequence Learning with Neural Networks
Sutskever, I., Vinyals, O., & Le, Q. V. (2014) · arXiv · doi:10.48550/arxiv.1409.3215 · Cited by 17,766 · cited 1×
“These methods have gained popularity over the last ten years thanks to empirical achievements on a wide range of tasks including speech recognition [2] [3], computer vision [4], and natural language processing [5] [6].”
BackgroundNeural Networks for Pattern Recognition
Bishop, C. M. (1995) · doi:10.1093/oso/9780198538493.001.0001 · Cited by 12,287
Hinton, G., Deng, L., Yu, D., Dahl, G., Mohamed, A.-R., Jaitly, N., Senior, A., Vanhoucke, V., Nguyen, P., Sainath, T., et al. (2012) · IEEE Signal Processing Magazine · doi:10.1109/msp.2012.2205597 · Cited by 10,260 · cited 1×
“These methods have gained popularity over the last ten years thanks to empirical achievements on a wide range of tasks including speech recognition [2] [3], computer vision [4], and natural language processing [5] [6].”
BackgroundLearning Deep Architectures for AI
Bengio, Y. (2009) · Foundations and Trends® in Machine Learning · doi:10.1561/2200000006 · Cited by 9,239 · cited 1×
“Deep neural networks (DNNs) are defined as networks with one or more hidden layers [61, 62].”
BackgroundMultilayer feedforward networks are universal approximators
HornikK., StinchcombeM., & WhiteH. (1989) · Neural Networks · doi:10.5555/70405.70408 · Cited by 9,081
(2011) · Cited by 6,993
Practical Bayesian Optimization of Machine Learning Algorithms
Snoek, J., Larochelle, H., & Adams, R. P. (2012) · arXiv · doi:10.48550/arxiv.1206.2944 · Cited by 6,691 · cited 1×
“There is currently no standard process for setting/tuning DNN hyperparameters, which remains somewhere between a black art and cutting-edge research [33].”
BackgroundA unified architecture for natural language processing: deep neural networks with multitask learning
Collobert, R., & Weston, J. (2008) · Proceedings of the 25th international conference on Machine learning - ICML '08 · doi:10.1145/1390156.1390177 · Cited by 5,058 · cited 1×
“These methods have gained popularity over the last ten years thanks to empirical achievements on a wide range of tasks including speech recognition [2] [3], computer vision [4], and natural language processing [5] [6].”
BackgroundLearning Deep Architectures for AI
Bengio, Y. (2009) · now publishers, Inc. eBooks · doi:10.1561/9781601982957 · Cited by 4,941
Convolutional networks for images, speech, and time series
LeCun, Y., & Bengio, Y. (1998) · HAL (Le Centre pour la Communication Scientifique Directe) · doi:10.5555/303568.303704 · Cited by 4,231
Networks of spiking neurons: The third generation of neural network models
Maass, W. (1997) · Neural Networks · doi:10.1016/s0893-6080(97)00011-7 · Cited by 3,481 · cited 1×
“Spiking Neural Networks (SNNs) [57, 58, 59, 60] are popular in ASIC designs, largely due to their compact mapping onto fully-parallel hardware.”
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 · cited 1×
“The reconfigurable CNN accelerator in [53] clock gates processing elements when activities are zero.”
Background(1989) · Cited by 3,214
(2013) · Cited by 2,056
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 · cited 1×
“In [52], the authors focus on compressing DNN weights, but also propose skipping activities of zero value, eliding up to 70% of the neuron computations.”
BackgroundDeep Learning with Limited Numerical Precision
Gupta, S., Agrawal, A., Gopalakrishnan, K., & Narayanan, P. (2015) · International Conference on Machine Learning · Cited by 1,590
ISAAC: a convolutional neural network accelerator with in-situ analog arithmetic in crossbars
Shafiee, A., Nag, A., Muralimanohar, N., Balasubramonian, R., Strachan, J. P., Hu, M., Williams, R. S., & Srikumar, V. (2016) · ACM SIGARCH Computer Architecture News · doi:10.1145/3007787.3001139 · Cited by 1,536 · cited 1×
“Others have proposed reducing computation and data movement costs with in-situ processing using memristors [50].”
BackgroundDaDianNao: 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×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Learning both Weights and Connections for Efficient Neural Networks
Han, S., Pool, J., Tran, J., & Dally, W. J. (2015) · arXiv · doi:10.48550/arxiv.1506.02626 · Cited by 1,481 · cited 1×
“Weight sparsity was considered in [51], where the authors co-design training and prediction to favor weight sparsity then statically prune the network to remove unimportant edges.”
BackgroundDeep Learning with Limited Numerical Precision
Gupta, S., Agrawal, A., Gopalakrishnan, K., & Narayanan, P. (2015) · arXiv · doi:10.48550/arxiv.1502.02551 · Cited by 1,364 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
DianNao: a small-footprint high-throughput accelerator for ubiquitous machine-learning
Chen, T., Du, Z., Sun, N., Wang, J., Wu, C., Chen, Y., & Temam, O. (2014) · Proceedings of the 19th international conference on Architectural support for programming languages and operating systems · doi:10.1145/2541940.2541967 · Cited by 1,297 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
A Probabilistic Analysis of the Rocchio Algorithm with TFIDF for Text Categorization
Joachims, T. (1997) · Cited by 1,262
Efficient Learning of Sparse Representations with an Energy-Based Model
Ranzato, M., Poultney, C., Chopra, S., & LeCun, Y. (2007) · Advances in Neural Information Processing Systems 19 · doi:10.7551/mitpress/7503.003.0147 · Cited by 1,106 · cited 1×
“References: CPUs [8, 9, 10, 11, 12], GPUs [13, 14, 15, 16, 8, 9, 11], FPGAs [17, 12], and ASICs [18, 19, 20, 21, 13, 22, 12, 23].”
Deep, Big, Simple Neural Nets for Handwritten Digit Recognition
Cireşan, D. C., Meier, U., Gambardella, L. M., & Schmidhuber, J. (2010) · Neural Computation · doi:10.1162/neco_a_00052 · Cited by 1,078 · cited 1×
“References: CPUs [8, 9, 10, 11, 12], GPUs [13, 14, 15, 16, 8, 9, 11], FPGAs [17, 12], and ASICs [18, 19, 20, 21, 13, 22, 12, 23].”
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) · 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) · doi:10.1109/isca.2016.30 · Cited by 1,077
GHOSH-DASTIDAR, S., & ADELI, H. (2009) · International Journal of Neural Systems · doi:10.1142/s0129065709002002 · Cited by 1,057
Neural Networks with Dynamic Synapses
Tsodyks, M., Pawelzik, K., & Markram, H. (1998) · Neural Computation · doi:10.1162/089976698300017502 · Cited by 942 · cited 1×
“Spiking Neural Networks (SNNs) [57, 58, 59, 60] are popular in ASIC designs, largely due to their compact mapping onto fully-parallel hardware.”
BackgroundISAAC: A Convolutional Neural Network Accelerator with In-Situ Analog Arithmetic in Crossbars
Shafiee, A., Nag, A., Muralimanohar, N., Balasubramonian, R., Strachan, J. P., Hu, M., Williams, R. S., & Srikumar, V. (2016) · 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) · doi:10.1109/isca.2016.12 · Cited by 775
(2011) · Cited by 713
14.5 Eyeriss: An energy-efficient reconfigurable accelerator for deep convolutional neural networks
Chen, Y.-H., Krishna, T., Emer, J., & Sze, V. (2016) · 2016 IEEE International Solid-State Circuits Conference (ISSCC) · doi:10.1109/isscc.2016.7418007 · Cited by 713
Learning to extract symbolic knowledge from the World Wide Web
Craven, M. W., DiPasquo, D., Freitag, D., McCallum, A., Mitchell, T. M., Nigam, K., & Slattery, S. (1998) · National Conference on Artificial Intelligence · Cited by 662
RazorII: In Situ Error Detection and Correction for PVT and SER Tolerance
Das, S., Tokunaga, C., Pant, S., Ma, W.-H., Kalaiselvan, S., Lai, K., Bull, D. M., & Blaauw, D. T. (2009) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2008.2007145 · Cited by 598 · cited 1×
“In fault-tolerant CPU applications a typical correction mechanism is to recompute using checkpoint and replay [39, 41, 42], a capability specific to the sophisticated control-plane logic of”
MethodEnerJ: approximate data types for safe and general low-power computation
Sampson, A., Dietl, W., Fortuna, E., Gnanapragasam, D., Ceze, L., & Grossman, D. (2011) · ACM SIGPLAN Notices · doi:10.1145/1993316.1993518 · Cited by 589
Convolutional Neural Network Committees for Handwritten Character Classification
Ciresan, D. C., Meier, U., Gambardella, L. M., & Schmidhuber, J. (2011) · 2011 International Conference on Document Analysis and Recognition · doi:10.1109/icdar.2011.229 · Cited by 536 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
A Self-Tuning DVS Processor Using Delay-Error Detection and Correction
Das, S., Roberts, D., Lee, S., Pant, S., Blaauw, D., Austin, T., Flautner, K., & Mudge, T. (2006) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2006.870912 · Cited by 396 · cited 1×
“A number of techniques have been described to detect faults arising from voltage scaling, including parity bits [36], Razor double-sampling [37, 38], Razor transitiondetection [39], and various canary circuits [40].”
MethodSelf-citationEnerJ: approximate data types for safe and general low-power computation
Sampson, A., Dietl, W., Fortuna, E., Gnanapragasam, D., Ceze, L., & Grossman, D. (2011) · Proceedings of the 32nd ACM SIGPLAN Conference on Programming Language Design and Implementation · doi:10.1145/1993498.1993518 · Cited by 390 · cited 1×
“More recently, with the revival of DNNs, fixed-point bitwidth reduction has been revisited as a method to reduce power and area for resource-constrained computing [47, 48, 13, 49, 21, 17].”
MethodLearning Real-World Stimuli in a Neural Network with Spike-Driven Synaptic Dynamics
Brader, J. M., Senn, W., & Fusi, S. (2007) · Neural Computation · doi:10.1162/neco.2007.19.11.2881 · Cited by 371 · cited 1×
“Spiking Neural Networks (SNNs) [57, 58, 59, 60] are popular in ASIC designs, largely due to their compact mapping onto fully-parallel hardware.”
BackgroundA 45 nm Resilient Microprocessor Core for Dynamic Variation Tolerance
Bowman, K. A., Tschanz, J. W., Lu, S.-L. L., Aseron, P. A., Khellah, M. M., Raychowdhury, A., Geuskens, B. M., Tokunaga, C., Wilkerson, C. B., Karnik, T., et al. (2011) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2010.2089657 · Cited by 266 · cited 1×
“In fault-tolerant CPU applications a typical correction mechanism is to recompute using checkpoint and replay [39, 41, 42], a capability specific to the sophisticated control-plane logic of”
MethodShao, Y. S., Reagen, B., Wei, G.-Y., & Brooks, D. (2014) · ACM SIGARCH Computer Architecture News · doi:10.1145/2678373.2665689 · Cited by 241
Self-citationPerformance and Scalability of GPU-Based Convolutional Neural Networks
Strigl, D., Kofler, K., & Podlipnig, S. (2010) · 2010 18th Euromicro Conference on Parallel, Distributed and Network-based Processing · doi:10.1109/pdp.2010.43 · Cited by 240 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Networks of spiking neurons: the third generation of neural network models
MaasWofgang (1997) · Transactions of The Society for Computer Simulation International · doi:10.5555/281543.281637 · Cited by 239
Hardware accelerated convolutional neural networks for synthetic vision systems
Farabet, C., Martini, B., Akselrod, P., Talay, S., LeCun, Y., & Culurciello, E. (2010) · Proceedings of 2010 IEEE International Symposium on Circuits and Systems · doi:10.1109/iscas.2010.5537908 · Cited by 231 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Process variation in embedded memories: failure analysis and variation aware architecture
Agarwal, A., Paul, B. C., Mukhopadhyay, S., & Roy, K. (2005) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2005.852159 · Cited by 175 · cited 1×
“Fault distributions corresponding to each SRAM voltage reduction step are modeled using Monte Carlo SPICE simulation with 10,000 samples, similar to the methodology of [28].”
MethodBull, D., Das, S., Shivashankar, K., Dasika, G. S., Flautner, K., & Blaauw, D. (2011) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2010.2079410 · Cited by 171 · cited 2×
“In fault-tolerant CPU applications a typical correction mechanism is to recompute using checkpoint and replay [39, 41, 42], a capability specific to the sophisticated control-plane logic of”
MethodDjiNN and Tonic: DNN as a service and its implications for future warehouse scale computers
Hauswald, J., Kang, Y., Laurenzano, M. A., Chen, Q., Li, C., Mudge, T., Dreslinski, R. G., Mars, J., & Tang, L. (2015) · Proceedings of the 42nd Annual International Symposium on Computer Architecture · doi:10.1145/2749469.2749472 · Cited by 161 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
ApproxANN: an approximate computing framework for artificial neural network
Zhang, Q., Wang, T., Tian, Y., Yuan, F., & Xu, Q. (2015) · Design, Automation, and Test in Europe · doi:10.5555/2755753.2755913 · Cited by 139
A defect-tolerant accelerator for emerging high-performance applications
Temam, O. (2012) · ACM SIGARCH Computer Architecture News · doi:10.1145/2366231.2337200 · Cited by 139 · cited 3×
“The conventional approach to using fixed-point types in DNNs is to select a single, convenient bitwidth (typically 16 bits [34, 19, 21]) and verify the prediction error is within some margin of the original floating-point implementation.”
BackgroundShao, Y. S., Reagen, B., Wei, G.-Y., & Brooks, D. M. (2014) · Proceedings - International Symposium on Computer Architecture/Proceedings · doi:10.1109/isca.2014.6853196 · Cited by 135
Self-citationBuilding block of a programmable neuromorphic substrate: A digital neurosynaptic core
Arthur, J. V., Merolla, P. A., Akopyan, F., Alvarez, R., Cassidy, A., Chandra, S., Esser, S. K., Imam, N., Risk, W., Rubin, D. B. D., et al. (2012) · The 2012 International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2012.6252637 · Cited by 131 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Cognitive computing systems: Algorithms and applications for networks of neurosynaptic cores
Esser, S. K., Andreopoulos, A., Appuswamy, R., Datta, P., Barch, D., Amir, A., Arthur, J., Cassidy, A., Flickner, M., Merolla, P., et al. (2013) · The 2013 International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2013.6706746 · Cited by 125 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Comparison of neural networks and discriminant analysis in predicting forest cover types
Dean, D. J., & Blackard, J. A. (1998) · Cited by 118
Back propagation simulations using limited precision calculations
Holt, J. L., & Baker, T. (2002) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.1991.155324 · Cited by 102 · cited 1×
“Early investigations into data type precision focused purely on the relationship between reduced precision and prediction error [45, 46].”
BackgroundStromatias, E., Neil, D., Galluppi, F., Pfeiffer, M., Liu, S.-C., & Furber, S. (2015) · 2015 International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2015.7280625 · Cited by 97 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Kim, J. K., Knag, P. C., Chen, T., & Zhang, Z. (2015) · Symposium on VLSI Circuits · doi:10.1109/vlsic.2015.7231323 · Cited by 90 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Whatmough, P. N., Das, S., & Bull, D. M. (2013) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2013.2284364 · Cited by 53 · cited 1×
“This is reminiscent of an approach studied for fault-tolerant DSP accelerators [43, 44].”
ResultSelf-citationCircuit-Level Timing Error Tolerance for Low-Power DSP Filters and Transforms
Whatmough, P. N., Das, S., Bull, D. M., & Darwazeh, I. (2012) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2012.2202930 · Cited by 51 · cited 1×
“This is reminiscent of an approach studied for fault-tolerant DSP accelerators [43, 44].”
ResultSelf-citationKung, J., Kim, D., & Mukhopadhyay, S. (2015) · 2015 IEEE/ACM International Symposium on Low Power Electronics and Design (ISLPED) · doi:10.1109/islped.2015.7273495 · Cited by 44 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Guest Editorial: Deep Learning
Ranzato, M., Hinton, G. E., & LeCun, Y. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0813-1 · Cited by 41 · cited 1×
“The recent success of DNNs in these problem domains can be attributed to three key factors: the availability of massive datasets, access to highly parallel computational resources such as GPUs, and improvements in the algorithms used to tune DNNs to data [7].”
BackgroundReduction of required precision bits for back-propagation applied to pattern recognition
Sakaue, S., Kohda, T., Yamamoto, H., Maruno, S., & Shimeki, Y. (1993) · IEEE Transactions on Neural Networks · doi:10.1109/72.207614 · Cited by 35 · cited 1×
“Early investigations into data type precision focused purely on the relationship between reduced precision and prediction error [45, 46].”
BackgroundDesign and Analysis of A 5.3-pJ 64-kb Gated Ground SRAM With Multiword ECC
Jahinuzzaman, S. M., Shah, J. S., Rennie, D. J., & Sachdev, M. (2009) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2009.2021088 · Cited by 26 · cited 1×
“A number of techniques have been described to detect faults arising from voltage scaling, including parity bits [36], Razor double-sampling [37, 38], Razor transitiondetection [39], and various canary circuits [40].”
MethodTunable Replica Bits for Dynamic Variation Tolerance in 8T SRAM Arrays
Raychowdhury, A., Geuskens, B. M., Bowman, K. A., Tschanz, J. W., Lu, S., Karnik, T., Khellah, M. M., & De, V. K. (2011) · IEEE Journal of Solid-State Circuits · doi:10.1109/jssc.2011.2108141 · Cited by 23 · cited 1×
“A number of techniques have been described to detect faults arising from voltage scaling, including parity bits [36], Razor double-sampling [37, 38], Razor transitiondetection [39], and various canary circuits [40].”
Retraining-based timing error mitigation for hardware neural networks
Deng, J., Fang, Y., Du, Z., Wang, Y., Li, H., Temam, O., Ienne, P., Novo, D., Li, X., Chen, Y., et al. (2015) · Design, Automation, and Test in Europe · doi:10.5555/2755753.2755887 · Cited by 19
Kulkarni, J. P., Tokunaga, C., Aseron, P. A., Nguyen, T., Augustine, C., Tschanz, J. W., & De, V. K. (2015) · Digest of technical papers/Digest of technical papers - IEEE International Solid-State Circuits Conference · doi:10.1109/isscc.2015.7062936 · Cited by 17 · cited 2×
“3% for power and area respectively [38].”
(2014) · doi:10.4230/oasics.slate.2021.16 · Cited by 1 · cited 1×
“These methods have gained popularity over the last ten years thanks to empirical achievements on a wide range of tasks including speech recognition [2] [3], computer vision [4], and natural language processing [5] [6].”
BackgroundUntitled work
Srivastava (2014) · doi:10.5555/2627435.2670313 · cited 1×
“References: CPUs [8] [9] [10] [11] [12], GPUs [13] [14] [15] [16] [8] [9] [11], FPGAs [17] [12], and ASICs [18] [19] [20] [21] [13] [22] [12] [23]. outputs from new input data.”
Research works printed but not linked: 2
- "Keras : Theano-based deep learning library." http://keras.io , 2015 . "Keras: Theano-based deep learning library." http://keras.io, 2015.
- A. L. Maas , A. Y. Hannun , and A. Y. Ng , " Rectifier nonlinearities improve neural network acoustic models," in ICML , vol. 30 , 2013 . A. L. Maas, A. Y. Hannun, and A. Y. Ng, "Rectifier nonlinearities improve neural network acoustic models," in ICML, vol. 30, 2013.
Cited by
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
Efficient Processing of Deep Neural Networks: A Tutorial and Survey
Sze, V., Chen, Y.-H., Yang, T.-J., & Emer, J. S. (2017) · Proceedings of the IEEE · doi:10.1109/jproc.2017.2761740 · Cited by 4,215
Edge Intelligence: Paving the Last Mile of Artificial Intelligence With Edge Computing
Zhou, Z., Chen, X., Li, E., Zeng, L., Luo, K., & Zhang, J. (2019) · Proceedings of the IEEE · doi:10.1109/jproc.2019.2918951 · Cited by 2,455
“[74] B. Reagen et al., “Minerva: Enabling low-power, highly-accurate deep neural network accelerators,” ACM SIGARCH Comput.”
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
“Minerva [39] also uses data quantization to save memory energy.”
MethodEIE: 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) · 2016 ACM/IEEE 43rd Annual International Symposium on Computer Architecture (ISCA) · doi:10.1109/isca.2016.30 · Cited by 1,077
Metadata
- Authors
- Reagen, B., Whatmough, P., Adolf, R., Rama, S., Lee, H., Lee, S. K., Hernández-Lobato, J. M., Wei, G.-Y., & Brooks, D.
- Affiliations
- Harvard University (US)
- ORCID
- Venue
- ACM SIGARCH Computer Architecture News
- Published
- 2016-06-18
- Volume
- 44
- Issue
- 3
- Pages
- 267-278
- ISSN
- 0163-5964 (ISSN-L)
- OpenAlex
- W2513554817
- Cited by
- 446
- References
- 70
- Citations by year
- 2026: 1 · 2025: 16 · 2024: 14 · 2023: 22 · 2022: 26 · 2021: 67 · 2020: 84 · 2019: 77 · 2018: 74 · 2017: 55 · 2016: 9
- FWCI
- 33.01
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- ACM SIGARCH Computer Architecture News (publishedVersion)
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Low-power high-performance VLSI design
- 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
- Keywords
- deep neural network accelerators, energy consumption reduction, heterogeneous data types, hardware fault detection, error mitigation, energy-efficient computing, IoT devices, mobile devices
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.