Pith. sign in

REVIEW 4 major objections 4 minor 67 references

MASR: A Modular Accelerator for Sparse RNNs

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A modular accelerator for speech RNNs can skip both zero weights and zero activations, achieving 14x speedups over dense baselines by computing sparse addresses in logic.

desk verdict Solid sparse-RNN accelerator paper; the architecture is credible, but the headline gains depend on unquantified model-level assumptions that the authors should be asked to release. read the letter →

arxiv 1908.08976 v1 pith:EVM56RF6 submitted 2019-08-23 eess.SP

Udit Gupta , Brandon Reagen , Lillian Pentecost , Marco Donato , Thierry Tambe , Alexander M. Rush , Gu-Yeon Wei , David Brooks This is my paper · ORCID
classification eess.SP
keywords recurrentneuralnetworksspeechrecognitionhardwareacceleratoractivationsparsityweightbitmasksparseencodingdynamicloadbalancingbidirectionalRNN
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

MASR argues that the main memory and compute bottleneck in RNN-based speech recognition is dynamic activations, not just weights, and that an accelerator can compress and skip both. The paper's core proposal is a microarchitecture that stores only non-zero weights and activations, uses binary masks plus small logic circuits to find matching non-zero pairs, and dynamically rebalances work across parallel lanes. Evaluated on a distilled, pruned, 10-bit quantized 5-layer bidirectional ReLU RNN, MASR is reported to give 14x performance, 2x area, and 15x energy benefits over a dense ASR RNN baseline, and 1.6x performance, 2x area, and 3x energy benefits over EIE, a prior sparse DNN accelerator. If those numbers hold, MASR would be an RNN accelerator that compresses and skips both static weights and dynamic activations while scaling from small 32-lane designs to 1024-lane designs.

What carries the argument

The central object is the work mask, formed by ANDing the weight mask with the activation mask; it represents exactly the set of non-zero weight-activation pairs that must be multiplied. A leading-non-zero detector (a logic circuit that finds the next set bit) steps through the work mask, and population counts on the two source masks generate addresses of compactly stored values, moving sparse-index complexity from memory to logic. The decoupled lane pipeline and the dynamic load balancer are the supporting machinery that keep those sparse operations from stalling or idling unevenly across the processing array.

What would settle it

Measure the distilled 5-layer ReLU RNN's word error rate on the LibriSpeech corpus used in the paper and count zeros through the network: if WER rises materially above the 21.9 GRU baseline, or if the average hidden-state zero fraction falls well below about 80% (and inputs below about 60%), the reported speedups and energy savings are measured on a less accurate or denser model.

Watch

Extended reading notes

Core claim

The paper's central claim is that a bitmask-based sparse encoding is sufficient to make sparse bidirectional RNN inference efficient at scale. Instead of storing row and column pointers as in compressed sparse row formats, MASR keeps a one-bit mask for weights and a one-bit mask for activations; a bitwise AND produces a work mask, a leading-non-zero detector finds the next pair of non-zero values, and population counts compute the compact memory addresses of those values in logic. This removes pointer storage that grows with the number of processing elements, compresses activations as well as weights, and ensures no MAC is issued for a null weight-activation pair. A decoupled front-end/back-end pipeline and a dynamic load balancer that redistributes non-zero activations at run time convert the sparse address stream into high MAC utilization. The measured claim is 14x performance, 2x area, and 15x energy versus a dense ASR RNN baseline, and 1.6x performance, 2x area, and 3x energy versus EIE, with the advantage over EIE widening beyond 128 parallel MACs.

Load-bearing premise

The load-bearing premise is that a distilled 5-layer bidirectional ReLU RNN matches the accuracy of the 5-layer GRU teacher and stays about 60% zero in inputs and 80% zero in hidden states after batch-norm refactoring and 10-bit quantization.

Editorial extensions

If this is right

  • An RNN accelerator can treat activation memory as a first-class compression target, not just weights, because batch-norm refactoring and ReLU-based distillation make roughly 60% of inputs and 80% of hidden states zero without losing accuracy.
  • Binary-mask sparse encoding makes the accelerator's memory overhead independent of the number of processing elements, so the design can scale to 1024 lanes where row-pointer-based accelerators see pointer storage dominate.
  • Dynamic load balancing over non-zero activations recovers up to 1.8x performance and lifts MAC utilization to roughly 90%, 80%, and 50% for the 64-, 256-, and 1024-lane configurations.
  • Double-buffering weights and activations lets the same modular design process speech samples of arbitrary length with DRAM energy remaining a small fraction of total energy.
  • If the reported comparisons hold, a sparse RNN accelerator can outperform both dense RNN execution and existing sparse-DNN accelerators on the same optimized model, with the gap growing as parallelism increases.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same bitmask-plus-work-mask approach should transfer to any recurrent layer that can be made activation-sparse; a direct test would train a ReLU-activated GRU or transducer encoder with the same distillation recipe and check whether the 60% input and 80% hidden-state zero fractions persist.
  • The synthetic scaling results imply that accuracy-preserving pruning methods reaching 10% non-zero weights and activations would yield roughly 26x energy savings and 76x speedups on larger RNNs, but those numbers assume the sparsity can be achieved at the same word error rate.
  • The full 1024-lane utilization claim rests on simulation plus one placed-and-routed 32-lane PE; measuring a fabricated or RTL-validated 1024-lane design would settle whether the dynamic load balancer and logic-computed addressing hold up at physical scale.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper presents MASR, a modular accelerator for bidirectional RNNs used in on-chip ASR. The design exploits sparsity in both weights and activations through a bitmask encoding that replaces pointer-based sparse formats with logic-computed addresses, combined with a decoupled pipelined lane architecture and dynamic load balancing to address irregularity. The authors construct an optimized RNN baseline via knowledge distillation, weight pruning, and 10-bit quantization, and they evaluate MASR across 32 to 1024 lanes using a cycle-level simulator validated against synthesized RTL within 10% power and 12% area error. The headline claims are a 1.6x performance, 2x area, and 3x energy benefit over EIE, and a 14x performance and 15x energy benefit over a dense ASR RNN baseline.

Significance. If the central claims hold, MASR is a significant contribution: it is among the first sparse RNN accelerators to exploit both dynamic activation sparsity and static weight sparsity, and its logic-centric encoding is a clean alternative to CSR formats for highly parallel designs. The hardware validation is a notable strength: the simulator is checked against placed-and-routed RTL with small error, and the authors report a fabricated SoC. The paper also makes a credible design-space argument, showing how the same modular fabric scales from small to large configurations. The significance is conditional, however, because the headline speedups rest on two premises that are under-documented: the distilled student RNN matching the teacher's WER, and the claimed 60%/80% activation sparsity after quantization. The comparison to EIE/ESE is also analytical rather than measured, so the relative claims carry additional uncertainty.

major comments (4)
  1. [Section III-C / Figure 4] The paper claims that distillation followed by 5 epochs of CTC fine-tuning yields a 5-layer ReLU RNN 'with the same accuracy as the teacher,' but it never reports the student's WER. The teacher's WER (21.9) is stated, while Figure 4 is only a qualitative bar chart, and no test split, confidence interval, or error bar is given. Because all of the reported speedups (14x performance, 15x energy, and the EIE-relative numbers) are measured on this student model, the iso-accuracy premise is load-bearing. Please report the student's WER on the same LibriSpeech evaluation set as the teacher, ideally with per-utterance or bootstrap uncertainty, and state whether the student WER is statistically indistinguishable from 21.9.
  2. [Section IV-A / Table III] The activation sparsity figures are presented as facts without measurement details: 'Training with ReLU causes 80% of the hidden state values to be zero' and 'After refactoring, inputs are on average 60% zeros.' No per-layer sparsity, no indication of whether the numbers are averages over timesteps, utterances, or layers, and no sensitivity to the 10-bit quantization are provided. These numbers feed directly into the performance and energy results in Section VII-B and Figure 8. Please report per-layer hidden-state and input sparsity measured on the actual test set after quantization, along with the variance across utterances, so that the reader can assess how representative the 60%/80% values are.
  3. [Section IX] The comparison to EIE and ESE is an analytical cost model built from assumed memory-access patterns, a 50% utilization estimate for EIE, and a claim that CSR adds 'up to 40% wasted work,' rather than a cycle-level simulation of those accelerators on the same RNN workload. The paper does not cite a source for the 50% utilization figure or the 40% wasted-work estimate. Since the abstract's 2x/3x/1.6x benefits relative to EIE are exactly these model-based estimates, the relative claims should either be supported by a faithful simulation of EIE/ESE (or a documented implementation), or clearly labeled as analytical estimates with a sensitivity analysis over the utilization and wasted-work assumptions. In its current form, the strength of the relative claims exceeds the evidence.
  4. [Abstract and Section X] The abstract states that MASR provides '2× area 3× energy' benefits relative to EIE, while the conclusion states 'performance, area, and energy by 1.6×, 3×, and 2×.' Section IX and Figure 15 support the abstract's ordering (approximately 2x area and 3x energy), so the conclusion has the area and energy numbers swapped. This is an inconsistency in the paper's central headline claim and must be corrected so that the abstract, body, and conclusion agree.
minor comments (4)
  1. [Section V-B] The sentence 'Separate accumulators are maintained for the positive and negative weights as they were quantized separately (see in Section 2)' refers to a nonexistent 'Section 2'; the reference should point to the quantization discussion in Section III-C.
  2. [Section VIII-A] The text contains the typo '0.3mm m2' before 'This comes at a negligible area penalty of 0.3mm2'; the first phrase should be removed or corrected.
  3. [Section VII-B] The output-predication mechanism (OP) is described in a single paragraph and is included in the performance breakdown as providing up to 15% benefit, but no implementation details, prediction accuracy, or area/energy overhead are given. Since it contributes to the reported speedups, please either add these details or clearly mark it as a simulated policy rather than a fully implemented path.
  4. [Section III-C and Section VII-B] The 'dense ASR RNN baseline' is defined only implicitly as a CPU running a dense 7-layer RNN; the first appearance of the 14x/15x claim in the introduction should state the baseline explicitly (CPU execution of the dense 7-layer bidirectional RNN) to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity found: the hardware performance, area, and energy claims are grounded in RTL-validated simulation and real trained-model sparsity, not in the conclusions themselves.

full rationale

MASR does not derive a formal mathematical result whose inputs are its outputs; it reports measured and simulated engineering outcomes. The central performance, area, and energy numbers come from a cycle-level simulator validated against placed-and-routed 16nm RTL within 10% power and 12% area, with a fabricated SoC mentioned in Section VIII-C. The workload sparsity figures (80% hidden-state zeros due to ReLU, 60% input zeros after batch-norm refactoring) are empirical properties of a trained 5-layer bidirectional RNN on LibriSpeech, not quantities defined so that the hardware speedup is true by construction. The comparisons to EIE and ESE in Section IX are analytic and based on assumed memory-access patterns and utilization, but that is a modeling-accuracy limitation rather than circularity, because the EIE/ESE estimates are not derived from MASR's own measured results. The paper cites some prior work by its own authors (e.g., Weightless [58]), but only as a pruning/compression technique and not as load-bearing evidence for the accelerator's claimed benefits. The lack of reported numeric WER for the distilled student and the absence of confidence intervals for sparsity are validation and reproducibility concerns, not evidence that a prediction reduces to its inputs. There is no self-citation chain, no uniqueness theorem imported from the authors, no ansatz smuggled in via citation, and no fitted parameter renamed as a prediction. The headline benefits are conditional on the trained network being iso-accurate and sparse, but that conditionality is an assumption about the external workload, not a circular derivation.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper introduces no speculative physical entities. Its quantitative claims depend on measured sparsity in a specific distilled RNN, on the validity of BN folding under quantization, and on the workload assumption that RNNs dominate ASR time. Design-space parameters such as lane counts, queue depth, and banks are swept, not fitted to produce the result.

assumptions (4)
  • domain assumption Recurrent layers, not the language model, dominate ASR execution time (Section III-C: with beam width 128, only 10% of CPU time is beam search).
    Justifies focusing the accelerator on RNN matrix-vector products.
  • domain assumption The trained ReLU RNN has 80% zero hidden states and 60% zero inputs after batch-norm folding (Section IV-A); these ratios are used to estimate speedups and energy savings.
    If real deployment sparsity differs, the acceleration factors change.
  • domain assumption Batch normalization can be mathematically folded into the next layer's weights and biases during inference (Section IV-A), remaining valid under 10-bit quantized fixed point.
    Folding is exact in floating point; the paper applies it before quantization without reporting a separate accuracy check after folding.
  • standard math Bitwise AND of weight and activation masks equals the set of required multiply-accumulate operations (Section IV-B).
    Standard Boolean identity, assuming exact zero representation in fixed point.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MASR: A Modular Accelerator for Sparse RNNs." pith.science (2026). https://pith.science/paper/EVM56RF6

@misc{pith2026190808976,
  author       = {Pith},
  title        = {Pith review of: MASR: A Modular Accelerator for Sparse RNNs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EVM56RF6}},
  note         = {Machine review of arXiv:1908.08976}
}
read the original abstract

Recurrent neural networks (RNNs) are becoming the de facto solution for speech recognition. RNNs exploit long-term temporal relationships in data by applying repeated, learned transformations. Unlike fully-connected (FC) layers with single vector matrix operations, RNN layers consist of hundreds of such operations chained over time. This poses challenges unique to RNNs that are not found in convolutional neural networks (CNNs) or FC models, namely large dynamic activation. In this paper we present MASR, a principled and modular architecture that accelerates bidirectional RNNs for on-chip ASR. MASR is designed to exploit sparsity in both dynamic activations and static weights. The architecture is enhanced by a series of dynamic activation optimizations that enable compact storage, ensure no energy is wasted computing null operations, and maintain high MAC utilization for highly parallel accelerator designs. In comparison to current state-of-the-art sparse neural network accelerators (e.g., EIE), MASR provides 2x area 3x energy, and 1.6x performance benefits. The modular nature of MASR enables designs that efficiently scale from resource-constrained low-power IoT applications to large-scale, highly parallel datacenter deployments.

Figures

Figures reproduced from arXiv: 1908.08976 by the authors.

Figure 1
Figure 1. The memory footprint of activations is higher in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Compared to compressed sparse row encoding (e.g., [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Bidirectional RNN layer. x t , h t , and g t are the input and hidden states at time step t. Wx, Wh, Vx, and Vh are the forward and backward weights. support for sparsity in weights and activations, and dynamic load balancing. Typically, previous work either exploits sparsity in weights or activations, but not both [10], [11], [13], [14], [22], [46], [47] leaving key performance, area, and energy savings on the tabl… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Left: After distillation, a 5-layer bidirectional RNN reaches the accuracy of a 5-layer bidirectional GRU reducing the number of parameters by 3×. Right: Language modeling reduces the WER from 22 down to 14.5 with a beam-width of 128. Language modeling accounts for onl…
Figure 6
Figure 6. Figure 6: Overall topology of how weight matrices, input [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 5
Figure 5. Figure 5: MASR’s sparse encoding compute sparse addresses for [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: MASR accelerator design highlighting the overall system architecture (left), a PE (center), and a lane (right). Blocks [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Left: Energy-performance and area-performance Pareto frontiers of accelerator designs, sweeping microarchitectural parameters shown in [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Impact of increasing parallelism (lanes) on SRAM area [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: The plots on the left summarizes area (top row), energy (middle row), and power (bottom row) tradeoffs for the [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Energy breakdown of LANESx256 running speech [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 12
Figure 12. Figure 12: Left: As we scale the number of activation SRAM banks in the LANESx1024 architecture, the cycles spent on VVAdd decreases. Center: Scaling the depth for back-end accumulator queue trades off stalls for idle cycles. Right: The impact of increasing the number of activat…
Figure 14
Figure 14. Figure 14: MASR LANESx32 placed-and-routed layout B. Scaling RNN Size and Sparsity Recent advances in the machine learning community allow training sparser networks without sacrificing accuracy [57], [66], [67]. This suggests further performance and energy improvements may be po…
Figure 15
Figure 15. Figure 15: Area normalized to LANESx32 (top) and Energy [PITH_FULL_IMAGE:figures/full_fig_p012_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 52 canonical work pages

  1. [1]

    What is automatic speech recognition (ASR)?

    Amazon, “What is automatic speech recognition (ASR)?.” https://developer.amazon.com/alexa-skills-kit/asr, 2018

    work page 2018

  2. [2]

    Google Duplex: An AI system for accomplishing real-world tasks over the phone

    “Google Duplex: An AI system for accomplishing real-world tasks over the phone.” https://ai.googleblog.com/2018/05/duplex-ai-system-for- natural-conversation.html, 2018

    work page 2018

  3. [3]

    Deep speech 2: End-to-end speech recognition in english and mandarin,

    D. Amodei, R. Anubhai, E. Battenberg, C. Case, J. Casper, B. Catanzaro, J. Chen, M. Chrzanowski, A. Coates, G. Diamos, E. Elsen, J. Engel, L. Fan, C. Fougner, T. Han, A. Y . Hannun, B. Jun, P. LeGresley, L. Lin, S. Narang, A. Y . Ng, S. Ozair, R. Prenger, J. Raiman, S. Satheesh, D. Seetapun, S. Sengupta, Y . Wang, Z. Wang, C. Wang, B. Xiao, D. Yogatama, J...

    arXiv 2015

  4. [4]

    Long short-term memory recurrent neural network architectures for large scale acoustic modeling,

    H. Sak, A. Senior, and F. Beaufays, “Long short-term memory recurrent neural network architectures for large scale acoustic modeling,” in Fifteenth annual conference of the international speech communication association, 2014

    2014

  5. [5]

    Fast and accurate recurrent neural network acoustic models for speech recognition,

    H. Sak, A. W. Senior, K. Rao, and F. Beaufays, “Fast and accurate recurrent neural network acoustic models for speech recognition,” in INTERSPEECH, 2015

    work page 2015

  6. [6]

    Google voice search: faster and more accurate

    “Google voice search: faster and more accurate.” https://ai.googleblog.com/2015/09/google-voice-search-faster-and- more.html, 2015

    work page 2015

  7. [7]

    Librispeech: an asr corpus based on public domain audio books,

    D. P. Vassil Panayotov, Guoguo Chen and S. Khudanpur, “Librispeech: an asr corpus based on public domain audio books,” ICASSP’15, 2015

    work page 2015

  8. [8]

    EIE: efficient inference engine on compressed deep neural network,

    S. Han, X. Liu, H. Mao, J. Pu, A. Pedram, M. A. Horowitz, and W. J. Dally, “EIE: efficient inference engine on compressed deep neural network,” CoRR, vol. abs/1602.01528, 2016

    arXiv 2016

Show all 67 references
  1. [9]

    SCNN: an accelerator for compressed-sparse convolutional neural networks,

    A. Parashar, M. Rhu, A. Mukkara, A. Puglielli, R. Venkatesan, B. Khailany, J. S. Emer, S. W. Keckler, and W. J. Dally, “SCNN: an accelerator for compressed-sparse convolutional neural networks,” CoRR, vol. abs/1708.04485, 2017

    2017 arXiv

  2. [10]

    Cambricon-X: An accelerator for sparse neural networks,

    S. Zhang, Z. Du, L. Zhang, H. Lan, S. Liu, L. Li, Q. Guo, T. Chen, and Y . Chen, “Cambricon-X: An accelerator for sparse neural networks,” in 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO), pp. 1–12, Oct 2016

    2016

  3. [11]

    Eyeriss: An Energy- Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks,

    Y .-H. Chen, T. Krishna, J. Emer, and V . Sze, “Eyeriss: An Energy- Efficient Reconfigurable Accelerator for Deep Convolutional Neural Networks,” in ISSCC, 2016

    2016

  4. [12]

    Distilling the knowledge in a neural network,

    G. E. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” CoRR, vol. abs/1503.02531, 2015

    2015 arXiv

  5. [13]

    E-PUR: An energy- efficient processing unit for recurrent neural networks,

    F. Silfa, G. Dot, J.-M. Arnau, and A. Gonzalez, “E-PUR: An energy- efficient processing unit for recurrent neural networks,” 2017

    2017

  6. [14]

    Sparsenn: An energy-efficient neural network accelerator exploiting input and output sparsity,

    J. Zhu, J. Jiang, X. Chen, and C.-Y . Tsui, “Sparsenn: An energy-efficient neural network accelerator exploiting input and output sparsity,” in 2018 Design, Automation & Test in Europe Conference & Exhibition (DATE) , pp. 241–244, IEEE, 2018

    2018

  7. [15]

    In-datacenter performance analysis of a tensor processing unit,

    N. P. Jouppi, C. Young, N. Patil, D. Patterson, G. Agrawal, R. Bajwa, S. Bates, S. Bhatia, N. Boden, A. Borchers, R. Boyle, P.-l. Cantin, C. Chao, C. Clark, J. Coriell, M. Daley, M. Dau, J. Dean, B. Gelb, T. V . Ghaemmaghami, R. Gottipati, W. Gulland, R. Hagmann, C. R. Ho, D. ...

    2017

  8. [16]

    Nvidia deep learning accelerator (NVDLA),

    NVIDIA, “Nvidia deep learning accelerator (NVDLA),” 2018

    2018

  9. [17]

    Applied machine learning at facebook: A datacenter infrastructure perspective,

    K. Hazzelwood, S. Bird, D. Brooks, S. Chintala, U. Diril, M. Fawzy, B. Jia, Y . Jia, A. Kalro, J. Law, K. Lee, J. Lu, P. Noordhuis, M. Smelyan- skiy, L. Xiong, and X. Wang, “Applied machine learning at facebook: A datacenter infrastructure perspective,” in Proceedings of the 2...

    2018

  10. [18]

    Fathom: Reference workloads for modern deep learning methods,

    B. R. G.-Y . W. Robert Adolf, Saketh Rama and D. Brooks, “Fathom: Reference workloads for modern deep learning methods,” IISWC’16, 2016

    2016

  11. [19]

    Deep learning inference in facebook data centers: Characterization, performance optimizations and hardware implications,

    J. Park, M. Naumov, P. Basu, S. Deng, A. Kalaiah, D. Khudia, J. Law, P. Malani, A. Malevich, S. Nadathur, et al., “Deep learning inference in facebook data centers: Characterization, performance optimizations and hardware implications,” arXiv preprint arXiv:1811.09886 , 2018

    2018 arXiv

  12. [20]

    Diannao: A small-footprint high-throughput accelerator for ubiquitous machine-learning,

    T. Chen, Z. Du, N. Sun, J. Wang, C. Wu, Y . Chen, and O. Temam, “Diannao: A small-footprint high-throughput accelerator for ubiquitous machine-learning,” in ASPLOS, 2014

    2014

  13. [21]

    Dadiannao: A machine-learning supercomputer,

    Y . Chen, T. Luo, S. Liu, S. Zhang, L. He, J. Wang, L. Li, T. Chen, Z. Xu, N. Sun, and O. Teman, “Dadiannao: A machine-learning supercomputer,” in MICRO, 2014

    2014

  14. [22]

    Minerva: Enabling Low- Power, Highly-Accurate Deep Neural Network Accelerators,

    B. Reagen, P. Whatmough, R. Adolf, S. Rama, H. Lee, S. K. Lee, J. M. Hernandez-Lobato, G.-Y . Wei, and D. Brooks, “Minerva: Enabling Low- Power, Highly-Accurate Deep Neural Network Accelerators,” in ISCA, 2016

    2016

  15. [23]

    Cnvlutin: Ineffectual-neuron-free deep neural network computing,

    J. Albericio, P. Judd, T. Hetherington, T. Aamodt, N. E. Jerger, and A. Moshovos, “Cnvlutin: Ineffectual-neuron-free deep neural network computing,” SIGARCH Comput. Archit. News , vol. 44, pp. 1–13, June 2016

    2016

  16. [24]

    Circnn: Accelerating and compressing deep neural networks using block- circulant weight matrices,

    C. Ding, S. Liao, Y . Wang, Z. Li, N. Liu, Y . Zhuo, C. Wang, X. Qian, Y . Bai, G. Yuan, X. Ma, Y . Zhang, J. Tang, Q. Qiu, X. Lin, and B. Yuan, “Circnn: Accelerating and compressing deep neural networks using block- circulant weight matrices,” in Proceedings of the 50th Annua...

    2017

  17. [25]

    Isaac: A convolutional neural network accelerator with in-situ analog arithmetic in crossbars,

    A. Shafiee, A. Nag, N. Muralimanohar, R. Balasubramonian, J. P. Strachan, M. Hu, R. S. Williams, and V . Srikumar, “Isaac: A convolutional neural network accelerator with in-situ analog arithmetic in crossbars,” in Proceedings of the 43rd International Symposium on Computer Arc...

    2016

  18. [26]

    Compressing DMA engine: Leveraging activation sparsity for training deep neural networks,

    M. Rhu, M. O’Connor, N. Chatterjee, J. Pool, and S. W. Keckler, “Compressing DMA engine: Leveraging activation sparsity for training deep neural networks,” CoRR, vol. abs/1705.01626, 2017

    2017 arXiv

  19. [27]

    Scaledeep: A scalable compute architecture for learning and evaluating deep networks,

    S. Venkataramani, A. Ranjan, S. Banerjee, D. Das, S. Avancha, A. Jagan- nathan, A. Durg, D. Nagaraj, B. Kaul, P. Dubey, et al., “Scaledeep: A scalable compute architecture for learning and evaluating deep networks,” ACM SIGARCH Computer Architecture News , vol. 45, no. 2, pp. ...

    2017

  20. [28]

    Prediction based execution on deep neural networks,

    M. Song, J. Zhao, Y . Hu, J. Zhang, and T. Li, “Prediction based execution on deep neural networks,” in 2018 ACM/IEEE 45th Annual International Symposium on Computer Architecture (ISCA) , pp. 752–763, IEEE, 2018

    2018

  21. [29]

    Prime: A novel processing-in-memory architecture for neural network computation in reram-based main memory,

    P. Chi, S. Li, C. Xu, T. Zhang, J. Zhao, Y . Liu, Y . Wang, and Y . Xie, “Prime: A novel processing-in-memory architecture for neural network computation in reram-based main memory,” in ACM SIGARCH Computer Architecture News, vol. 44, pp. 27–39, IEEE Press, 2016

    2016

  22. [30]

    Stripes: Bit-serial deep neural network computing,

    P. Judd, J. Albericio, T. Hetherington, T. M. Aamodt, and A. Moshovos, “Stripes: Bit-serial deep neural network computing,” in 2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) , pp. 1–12, IEEE, 2016

    2016

  23. [31]

    Bit fusion: Bit-level dynamically composable architecture for accelerating deep neural networks,

    H. Sharma, J. Park, N. Suda, L. Lai, B. Chau, V . Chandra, and H. Es- maeilzadeh, “Bit fusion: Bit-level dynamically composable architecture for accelerating deep neural networks,” in Proceedings of the 45th Annual International Symposium on Computer Architecture , pp. 764–775...

    2018

  24. [32]

    Admm-nn: An algorithm-hardware co-design framework of dnns using alternating direction methods of multipliers,

    A. Ren, T. Zhang, S. Ye, J. Li, W. Xu, X. Qian, X. Lin, and Y . Wang, “Admm-nn: An algorithm-hardware co-design framework of dnns using alternating direction methods of multipliers,” in Proceedings of the Twenty-Fourth International Conference on Architectural Support for Prog...

    2019

  25. [33]

    Packing sparse convolutional neural networks for efficient systolic array implementations: Column combining under joint optimization,

    H. Kung, B. McDanel, and S. Q. Zhang, “Packing sparse convolutional neural networks for efficient systolic array implementations: Column combining under joint optimization,” in Proceedings of the Twenty-Fourth International Conference on Architectural Support for Programming La...

    2019

  26. [34]

    Split-cnn: Splitting window-based operations in convolutional neural networks for memory system optimization,

    T. Jin and S. Hong, “Split-cnn: Splitting window-based operations in convolutional neural networks for memory system optimization,” in Proceedings of the Twenty-Fourth International Conference on Archi- tectural Support for Programming Languages and Operating Systems , pp. 835...

    2019

  27. [35]

    E-rnn: Design optimization for efficient recurrent neural networks in fpgas,

    Z. Li, C. Ding, S. Wang, W. Wen, Y . Zhuo, C. Liu, Q. Qiu, W. Xu, X. Lin, X. Qian, et al., “E-rnn: Design optimization for efficient recurrent neural networks in fpgas,” in 2019 IEEE International Symposium on High Performance Computer Architecture (HPCA) , pp. 69–80, IEEE, 2019

    2019

  28. [36]

    A configurable cloud-scale dnn processor for real-time ai,

    J. F. K. O. M. Papamichael, T. M. M. Liu, D. L. S. A. M. Haselman, L. A. M. Ghandi, S. H. P. P. A. Sapek, and G. W. L. Woods, “A configurable cloud-scale dnn processor for real-time ai,” 13

  29. [37]

    Towards memory friendly long-short term memory networks (lstms) on mobile gpus,

    X. Zhang, C. Xie, J. Wang, W. Zhang, and X. Fu, “Towards memory friendly long-short term memory networks (lstms) on mobile gpus,” in 2018 51st Annual IEEE/ACM International Symposium on Microarchi- tecture (MICRO), pp. 162–174, IEEE, 2018

    2018

  30. [38]

    Maeri: Enabling flexible dataflow mapping over dnn accelerators via reconfigurable interconnects,

    H. Kwon, A. Samajdar, and T. Krishna, “Maeri: Enabling flexible dataflow mapping over dnn accelerators via reconfigurable interconnects,” in ACM SIGPLAN Notices, vol. 53, pp. 461–475, ACM, 2018

    2018

  31. [39]

    Astra: Exploiting predictability to optimize deep learning,

    M. Sivathanu, T. Chugh, S. S. Singapuram, and L. Zhou, “Astra: Exploiting predictability to optimize deep learning,” in Proceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS ’19, (New York, NY...

    2019

  32. [40]

    Tangram: Optimized coarse-grained dataflow for scalable nn accelerators,

    M. Gao, X. Yang, J. Pu, M. Horowitz, and C. Kozyrakis, “Tangram: Optimized coarse-grained dataflow for scalable nn accelerators,” in Pro- ceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems , pp. 807– 8...

    2019

  33. [41]

    Fpga-based accelerator for long short-term memory recurrent neural networks,

    Y . Guan, Z. Yuan, G. Sun, and J. Cong, “Fpga-based accelerator for long short-term memory recurrent neural networks,” in 2017 22nd Asia and South Pacific Design Automation Conference (ASP-DAC) , pp. 629–634, Jan 2017

    2017

  34. [42]

    Recurrent neural networks hardware implementation on fpga,

    E. C. Andre Xian Ming Chang, Berin Martini, “Recurrent neural networks hardware implementation on fpga,” 2016

    2016

  35. [43]

    Deltarnn: A power-efficient recurrent neural network accelerator,

    C. Gao, D. Neil, E. Ceolini, S.-C. Liu, and T. Delbruck, “Deltarnn: A power-efficient recurrent neural network accelerator,” in Proceedings of the 2018 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, FPGA ’18, (New York, NY , USA), pp. 21–30, ACM, 2018

    2018

  36. [44]

    Unfold: A memory- efficient speech recognizer using on-the-fly wfst composition,

    R. Yazdani, J.-M. Arnau, and A. Gonz ´alez, “Unfold: A memory- efficient speech recognizer using on-the-fly wfst composition,” in Proceedings of the 50th Annual IEEE/ACM International Symposium on Microarchitecture, MICRO-50 ’17, (New York, NY , USA), pp. 69–81, ACM, 2017

    2017

  37. [45]

    Deep speech: Scaling up end-to-end speech recognition,

    A. Y . Hannun, C. Case, J. Casper, B. Catanzaro, G. Diamos, E. Elsen, R. Prenger, S. Satheesh, S. Sengupta, A. Coates, and A. Y . Ng, “Deep speech: Scaling up end-to-end speech recognition,” CoRR, vol. abs/1412.5567, 2014

    2014 arXiv

  38. [46]

    Efficient processing of deep neural networks: A tutorial and survey,

    V . Sze, Y .-H. Chen, T.-J. Yang, and J. S. Emer, “Efficient processing of deep neural networks: A tutorial and survey,” Proceedings of the IEEE , vol. 105, no. 12, pp. 2295–2329, 2017

    2017

  39. [47]

    ESE: efficient speech recognition engine with compressed LSTM on FPGA,

    S. Han, J. Kang, H. Mao, Y . Hu, X. Li, Y . Li, D. Xie, H. Luo, S. Yao, Y . Wang, H. Yang, and W. J. Dally, “ESE: efficient speech recognition engine with compressed LSTM on FPGA,” CoRR, vol. abs/1612.00694, 2016

    2016 arXiv

  40. [48]

    Acoustic modeling using deep belief networks,

    A.-R. Mohamed, G. E. Dahl, and G. Hinton, “Acoustic modeling using deep belief networks,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 20, no. 1, pp. 14–22, 2012

    2012

  41. [49]

    Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,

    G. Hinton, L. Deng, D. Yu, G. E. Dahl, A.-r. Mohamed, N. Jaitly, A. Senior, V . Vanhoucke, P. Nguyen, T. N. Sainath, et al., “Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,” IEEE Signal Processing Magazine , vol. 29,...

    2012

  42. [50]

    Empirical evaluation of gated recurrent neural networks on sequence modeling,

    J. Chung, C. Gulcehre, K. Cho, and Y . Bengio, “Empirical evaluation of gated recurrent neural networks on sequence modeling,” arXiv preprint arXiv:1412.3555, 2014

    2014 arXiv

  43. [51]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  44. [52]

    A broad ml benchmark suite for measuring performance of ml software frameworks, ml hardware accelerators, and ml cloud platforms

    “A broad ml benchmark suite for measuring performance of ml software frameworks, ml hardware accelerators, and ml cloud platforms.” https://mlperf.org/, 2018

    2018

  45. [53]

    Pytorch

    “Pytorch.” http://pytorch.org/, 2017

    2017

  46. [54]

    deepspeech.pytorch

    “deepspeech.pytorch.” https://github.com/SeanNaren/deepspeech.pytorch, 2018

    2018

  47. [55]

    Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks,

    A. Graves, S. Fernandez, F. Gomez, and J. Schmidhuber, “Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks,” ICML’2006, 2006

    2006

  48. [56]

    The dark side of dnn pruning,

    R. Yazdani, M. Riera, J.-M. Arnau, and A. Gonzalez, “The dark side of dnn pruning,” in ISCA, 2018

    2018

  49. [57]

    Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding,

    S. Han, H. Mao, and W. J. Dally, “Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding,” CoRR, vol. abs/1510.00149, 2015

    2015 arXiv

  50. [58]

    Weightless: Lossy weight encoding for deep neural network compression,

    B. Reagen, U. Gupta, R. Adolf, M. M. Mitzenmacher, A. M. Rush, G.-Y . Wei, and D. Brooks, “Weightless: Lossy weight encoding for deep neural network compression,” arXiv preprint arXiv:1711.04686 , 2017

    2017 arXiv

  51. [59]

    Exploring sparsity in recurrent neural networks,

    S. Narang, E. Elsen, G. Diamos, and S. Sengupta, “Exploring sparsity in recurrent neural networks,” arXiv preprint arXiv:1704.05119 , 2017

    2017 arXiv

  52. [60]

    Exploring neural transducers for end-to-end speech recognition,

    E. Battenberg, J. Chen, R. Child, A. Coates, Y . G. Y . Li, H. Liu, S. Satheesh, A. Sriram, and Z. Zhu, “Exploring neural transducers for end-to-end speech recognition,” in 2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , pp. 206–213, IEEE, 2017

    2017

  53. [61]

    Streaming end-to-end speech recognition for mobile devices,

    Y . He, T. N. Sainath, R. Prabhavalkar, I. McGraw, R. Alvarez, D. Zhao, D. Rybach, A. Kannan, Y . Wu, R. Pang, et al., “Streaming end-to-end speech recognition for mobile devices,” arXiv preprint arXiv:1811.06621, 2018

    2018 arXiv

  54. [62]

    Monotonic chunkwise attention,

    C.-C. Chiu and C. Raffel, “Monotonic chunkwise attention,” arXiv preprint arXiv:1712.05382, 2017

    2017 arXiv

  55. [63]

    Deepbench

    “Deepbench.” https://github.com/baidu-research/DeepBench, 2018

    2018

  56. [64]

    SnaPEA : Predictive early activation for reducing computa- tion in deep convolutional neural networks,

    V . Akhlaghi, A. Yazdanbakhsh, K. Samadi, R. K. Gupta, and H. Es- maeilzadeh, “SnaPEA : Predictive early activation for reducing computa- tion in deep convolutional neural networks,” in Proceedings of the 45th International Symposium on Computer Architecture , 2018

    2018

  57. [65]

    Drampower: Open-source dram power and energy estimation tool

    K. Chandrasekar, C. Weis, Y . Li, S. Goossens, M. Jung, O. Naji, B. Akesson, N. Wehn, , and K. Goossens, “Drampower: Open-source dram power and energy estimation tool.” http://www.drampower.info

  58. [66]

    Soft weight-sharing for neural network compression,

    K. Ullrich, E. Meeds, and M. Welling, “Soft weight-sharing for neural network compression,” ICLR’2017, vol. abs/1702.04008, 2017

    2017 arXiv

  59. [67]

    Dynamic network surgery for efficient dnns,

    Y . Guo, A. Yao, and Y . Chen, “Dynamic network surgery for efficient dnns,” NIPS’2016, vol. abs/1608.04493, 2016. 14

    2016 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.