Pith. sign in

REVIEW 3 major objections 6 minor 34 references

QForce-RL: Quantized FPGA-Optimized Reinforcement Learning Compute Engine

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A quantized FPGA engine for reinforcement-learning policies is claimed to reach up to 2.6x the frame rate of prior accelerators at a fraction of the energy.

desk verdict A credible incremental FPGA RL accelerator whose headline speedups and reward-preservation claims are not supported by the evidence as presented. read the letter →

arxiv 2506.07046 v1 pith:CEWQGVVP submitted 2025-06-08 cs.AR cs.CVcs.ROeess.IV

Anushka Jha , Tanushree Dewangan , Mukul Lokhande , Santosh Kumar Vishvakarma This is my paper
classification cs.ARcs.CVcs.ROeess.IV
keywords reinforcementlearningFPGAacceleratorquantizationSIMDprocessingelementmultiply-accumulateunitCORDICactivationfunctionhierarchicaldeepedgeAI
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

The paper claims that reinforcement-learning policy inference can be made cheap enough for edge FPGAs by combining quantized fixed-point arithmetic with a configurable SIMD multiply-accumulate unit and a CORDIC-based activation unit. QForce-RL is built from these units into a hierarchical CNN/LSTM agent, and the reported results are 2.3x throughput and up to 2.6x frame rate over prior accelerators, with frame energies of 0.15 mJ for the fully connected configuration and 0.26 mJ for the LSTM configuration. The payoff, if the claims hold, is an FPGA design that lets resource-constrained devices run RL agents in real time without the memory and power cost of 32-bit floating-point computing. The accuracy evidence is the paper's load-bearing link: reward preservation is taken from a software quantization study, while the hardware itself is demonstrated through synthesis, emulation, and resource tables.

What carries the argument

The load-bearing mechanisms are Q-MAC and V-ACT. Q-MAC is a runtime-configurable SIMD multiply-accumulate unit that reuses sixteen 8-bit multipliers to deliver 16/4/1 MACs per cycle at FxP8/16/32 precision, shrinking LUT and flip-flop counts while keeping throughput up. V-ACT is a unified activation unit that derives ReLU, sigmoid, tanh, and softmax from low-latency CORDIC hyperbolic and linear stages, with an HOAA path that folds exponent handling into the hyperbolic output for a reported 21% gain. Together the two units let one quantized datapath serve conv, FC, and LSTM layers at three precisions, which is what turns quantization from a software choice into a hardware resource and energy saving.

What would settle it

Run the quantized policy network on the FPGA in a closed-loop RL environment, such as an Atari or continuous-control task from the software study, and compare cumulative reward against the FP32 baseline; a materially larger reward drop, or FPS/energy numbers that do not reproduce the reported 2.6x and 0.15 mJ figures on the same board, would refute the central claim.

Watch

Extended reading notes

Core claim

QForce-RL is presented as a complete FPGA-oriented compute engine for hierarchical deep RL in which convolution, fully connected, LSTM gating, and softmax all run on quantized fixed-point hardware with selectable 8/16/32-bit precision. The core datapath is a Q-MAC array of sixteen 8-bit multipliers that can be fused into four 16-bit or one 32-bit MAC per cycle, so a single unit delivers SIMD parallelism at low precision without a multiplier per lane. Activations go through V-ACT, a low-latency CORDIC-based unit that computes ReLU, sigmoid, tanh, and softmax and converges in (3n/8+1) cycles; the paper reports a 21% performance gain from its HOAA mechanism and up to 15% resource savings over an existing multi-precision SIMD processing element. In an 8-PE configuration at 232 MHz the authors report 11 GOPS and 2.8 GOPS for Q-FC and Q-LSTM, energy efficiencies of 26.1 and 7.8 GOPS/W, and 0.15/0.26 mJ per frame. The conclusion that RL performance is preserved rests on a software quantization study; the hardware evidence is synthesis and emulation, not closed-loop RL episodes.

Load-bearing premise

The central accuracy claim depends on the assumption that the rounding and activation approximations built into the hardware change the policy's reward by only the negligible amount seen in the software quantization study; the paper's reward curves come from that software study, not from running the FPGA or its RTL simulation.

Editorial extensions

If this is right

  • At 8-bit precision the synthesized 28nm Q-MAC reaches 6.37 TOPS/W and 6.28 TOPS/mm$^2$, about 3.8x the energy efficiency of the nearest multi-precision MAC baseline in the comparison table.
  • Replacing the standard multiplier inside Q-MAC with approximate alternatives such as CORDIC-mult, Quant-MAC, ILM, or posit-mult keeps quality of results at 98.4–99.2% and saves up to 42% area, so the accelerator can trade accuracy for resource use at deployment time.
  • The same Q-MAC and V-ACT blocks, assembled into a DNN accelerator, are claimed to yield 112 ms inference latency (69.3 GOPS) on a representative workload, beating three published edge accelerators with about 1% accuracy difference.
  • Parameterized PE count and precision selection let one architecture span FPGA and mobile CPU deployments, with 8-bit SIMD giving 2.6x latency improvement over 32-bit on a CPU core.

Reading between the lines

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

  • If the hardware really preserves the software-measured rewards, the Q-MAC/V-ACT pair could be reused in existing CNN/LSTM accelerators to give edge robots and drones real-time RL inference without retraining.
  • The paper validates inference only; an online training check would test whether quantized actors also preserve learning convergence when the accelerator interacts with the environment, not just when it replays a stored policy.
  • Because V-ACT covers softmax and tanh/sigmoid and Q-MAC is precision-scalable, the unit is not RL-specific, and the reported DNN numbers suggest the larger payoff may be a mixed-precision DNN accelerator.
  • The FPS gains are reported against different baselines in different tables; a direct same-board comparison against the hierarchical 32-bit baseline would separate the contribution of quantization from that of SIMD parallelism.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes QForce-RL, an FPGA-oriented quantized reinforcement-learning inference engine built from two hardware blocks: a configurable SIMD multiply-accumulate unit (Q-MAC) supporting FxP8/16/32 precisions and a CORDIC-based activation unit (V-ACT) supporting ReLU, Sigmoid, Tanh, and Softmax. The authors claim up to 2.3x performance enhancement and 2.6x better FPS compared with state-of-the-art accelerators, with no significant degradation in RL reward. The evidence consists of FPGA synthesis and post-implementation resource/power numbers for the Q-MAC and V-ACT, a Table V comparison of FPS/throughput/energy across several prior designs, CPU emulation latency results, and software quantization results taken from prior work (QuaRL).

Significance. If the empirical claims were properly supported, this work would address a real need: low-cost, reconfigurable compute elements tailored to RL workloads, where multi-precision SIMD and diverse activation functions matter. The proposed Q-MAC and V-ACT are plausible building blocks, and the reported resource numbers are modest. However, the central claims are not established by the evidence in the manuscript. The reward-preservation claim is borrowed from software quantization results, not from the proposed hardware datapath, and the headline speedups come from uncontrolled comparisons across different FPGAs and different application workloads. The paper's contribution is therefore best viewed as a preliminary hardware proposal rather than a validated RL compute engine.

major comments (3)
  1. [Section II, Fig. 3, and Section III-C] The claim that QForce-RL operates "without significant performance degradation" is not supported by any RL experiment on the proposed hardware. The reward plots in Fig. 3 are inherited from QuaRL [1] and concern software quantization, not the Q-MAC/V-ACT datapath with CORDIC approximations. Section III-C reports only FPGA synthesis, post-implementation resource/power numbers, and CPU emulation; no RTL simulation or on-FPGA RL episode is presented. Since quantization and CORDIC errors propagate through LSTM recurrent state, softmax, and environment feedback, the unit-level QoR values in Section III-A are insufficient evidence for reward preservation. Please add an end-to-end RL evaluation, such as RTL simulation or FPGA-in-the-loop with standard RL environments, to substantiate this central claim.
  2. [Table V and Section III-C] The headline 2.6x FPS and related performance comparisons are not controlled. The proposed Q-FC/Q-LSTM results are on Virtex-7, whereas E2HRL [2] is on Artix-7 and other baselines use Zynq, Arria 10, KCU15, ZCU102, or Nexys A7-100T; the workloads also differ (object detection versus RL). This contradicts the statement in Section III-C that "We have also re-implemented SoTA designs with the similar parameters for a fair comparison." For a valid comparison, the baselines should be re-implemented on the same FPGA family, with the same bit-width, and using the same RL policy and environment.
  3. [Section III-A] The approximate multiplier alternatives (ILM, CORDIC-mult, Posit-mult, Quant-MAC) are evaluated only by unit-level QoR between 98.4% and 99.2% on MAC operations. This is not evidence that end-to-end RL reward is preserved, because errors can accumulate in recurrent inference, interact with the environment, and affect policy quality. The claim of a "modest 1.8% accuracy loss" must be re-measured at the policy or episode level, not inferred from per-multiplier QoR.
minor comments (6)
  1. [Abstract and Table V] The claim of "performance enhancement up to 2.3x" is ambiguous; Table V shows 2.8x throughput for Q-FC versus E2HRL and 2.6x FPS, while 2.3x appears closest to the energy-efficiency ratio (26.1/11.4). Please state explicitly which metric the 2.3x refers to.
  2. [Throughout] The manuscript contains numerous typos and grammatical errors, including "became" in the first sentence, "efforts efforts" in Section I, and "Rge energy consumption" in Section III-C. A careful proofreading pass is needed.
  3. [Fig. 3(b)] The caption "Off-chip Components" does not match the plot, which shows convergence-time speedup across RL tasks. Please correct the caption or the figure content.
  4. [Section II, Eq. (1)] The uniform affine quantization formula is not clearly defined: specify whether the min/max are taken over the weight tensor, define the scale and zero-point notation, and clarify the meaning of Qn and Wq.
  5. [Section III-B] The convergence-cycle comparison "3n/8+1 cycles compared to (n/2+1) stages" is stated without derivation or a supporting reference. Please provide the basis for this claim.
  6. [Table III] The ASIC synthesis results should state the synthesis corner (SS/TT/FF), supply voltage, and cell library version for the TSMC 28nm technology, so that power and area numbers are reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: QForce-RL's throughput/resource claims come from fresh FPGA synthesis; the QuaRL-based reward-preservation assertion is an external-evidence extrapolation, not a circular reduction.

full rationale

The system-level claims that could be called predictions—FPS, throughput, power, energy per frame in Table V—are post-implementation FPGA synthesis/emulation results for the proposed Q-MAC and V-ACT stages (Section III-C), not quantities computed from an equation that assumes the claimed speedup. The paper's quantization benefit is inherited from QuaRL [1] and its architecture from E2HRL [2], but those are cited as external prior results, and the paper does not define QForce-RL's own outputs in terms of those citations. The self-cited component works (Flex-PE [11], Quant-MAC [17], LPRE [13], CORDIC-Neuron [24]) are used as design inspiration and as comparison baselines, not as an unverified uniqueness theorem or as the sole proof of the new hardware numbers. The main weakness is an evidence gap rather than circularity: Figure 3 reward curves come from QuaRL's software emulation, and no RL episode is run on the FPGA or RTL, so the 'without significant performance degradation' claim is extrapolated rather than demonstrated for this exact datapath. That is a validation limitation, not a self-referential derivation. No fitted parameter is renamed as a prediction, and no known result is simply relabeled as a new contribution.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claims rest on design choices (PE count, SIMD width), on quantized RL accuracy borrowed from QuaRL, and on an implicit assumption that different FPGAs are comparable. No new physical entities are introduced.

free parameters (2)
  • Number of PEs = 8 (peak config)
    The best configuration with 8 PEs achieves 11 GOPS at 196 mW; the FPS and energy numbers in Table V correspond to this chosen configuration.
  • SIMD lane width = 16 for FxP8, 4 for FxP16, 1 for FxP32
    The Q-MAC uses 16 parallel 8-bit multipliers to support different precisions; the throughput claims depend on this design choice.
assumptions (3)
  • domain assumption Uniform affine quantization (Eq. 1) preserves RL reward within negligible loss
    The paper relies on QuaRL [1] results (Fig. 3) rather than measuring hardware RL performance.
  • domain assumption CORDIC-based approximate activation functions maintain QoR between 98.4% and 99.2%
    Claimed in Section III-A without a dedicated on-hardware validation.
  • ad hoc to paper Virtex-7 and Artix-7 results are comparable for SoTA benchmarking
    Table V compares the proposed design on Virtex-7 with E2HRL on Artix-7, with no resource scaling or normalization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of QForce-RL: Quantized FPGA-Optimized Reinforcement Learning Compute Engine." pith.science (2026). https://pith.science/paper/CEWQGVVP

@misc{pith2026250607046,
  author       = {Pith},
  title        = {Pith review of: QForce-RL: Quantized FPGA-Optimized Reinforcement Learning Compute Engine},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CEWQGVVP}},
  note         = {Machine review of arXiv:2506.07046}
}
read the original abstract

Reinforcement Learning (RL) has outperformed other counterparts in sequential decision-making and dynamic environment control. However, FPGA deployment is significantly resource-expensive, as associated with large number of computations in training agents with high-quality images and possess new challenges. In this work, we propose QForce-RL takes benefits of quantization to enhance throughput and reduce energy footprint with light-weight RL architecture, without significant performance degradation. QForce-RL takes advantages from E2HRL to reduce overall RL actions to learn desired policy and QuaRL for quantization based SIMD for hardware acceleration. We have also provided detailed analysis for different RL environments, with emphasis on model size, parameters, and accelerated compute ops. The architecture is scalable for resource-constrained devices and provide parametrized efficient deployment with flexibility in latency, throughput, power, and energy efficiency. The proposed QForce-RL provides performance enhancement up to 2.3x and better FPS - 2.6x compared to SoTA works.

Figures

Figures reproduced from arXiv: 2506.07046 by the authors.

Figure 2
Figure 2. Q-Actor Software emulation setup (RL) reconfigurable SIMD processing element (PE) must efficiently support a wide range of precision levels for RL workloads as well, to enable high throughput at low energy per operation for resource-constrained AI deployments. Q-Actor [1] is a novel software-emulated reinforcement learning (RL) framework designed to accelerate distributed actor-learner training. It leveraged INT 8-b… view at source ↗
Figure 3
Figure 3. (a) Application performance comparison (rewards) for [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Q-HRL network architecture tion in reward performance, enabling significant end-to-end speedups across various RL algorithms as shown in Fig 3b. There is scope for exploring mixed-precision quantization, where different components of the RL workload such as the policy network, value estimator, or embedding layers operate at varying bit-widths. This approach can further op￾timize memory usage, minimize data transfer,… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: QForce-RL architecture frozen, and the sub-goal module is fine-tuned independently. The sub-goal path employs an MLP-style Q-FC block or a Q￾LSTM layer to model temporal dependencies. The Q-LSTM, a specialized RNN variant [11], retains long- and short-term context crit…
Figure 7
Figure 7. Figure 7: low-latency V-ACT architecture. multiplier in the Q-MAC unit with approximate alternatives : such as Iterative Log Multiplier (ILM) [23], CORDIC-based multiplier [24], Posit-mult [13], and Quant-MAC [17]can maintain a quality of results (QoR) between 98.4% and 99.2%, w…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 26 canonical work pages

  1. [1]

    QuaRL: Quantization for fast and environmentally sustainable reinforcement learning,

    S. Krishnan, M. Lam, et al. , “QuaRL: Quantization for fast and environmentally sustainable reinforcement learning,” IEEE Trans. on MLR, 2022

    work page 2022

  2. [2]

    E2HRL: An energy-efficient hardware ac- celerator for hierarchical deep reinforcement learning,

    A. Shiri, U. Kallakuri, et al., “E2HRL: An energy-efficient hardware ac- celerator for hierarchical deep reinforcement learning,” ACM TODAES, vol. 27, no. 5, 2022

    work page 2022

  3. [3]

    ChipNEMO: Domain-adapted LLMs for chip design,

    M. Liu, T.-D. Ene, et al., “ChipNEMO: Domain-adapted LLMs for chip design,” ACM CoRL, 2023

    work page 2023

  4. [4]

    Chip placement with deep reinforce- ment learning,

    A. Mirhoseini, A. Goldie, et al., “Chip placement with deep reinforce- ment learning,” arXiv preprint arXiv:2004.10746 , 2020

    arXiv 2004

  5. [5]

    Mastering the game of go without human knowledge,

    D. Silver, J. Schrittwieser, , et al. , “Mastering the game of go without human knowledge,” nature, vol. 550, 2017

    work page 2017

  6. [6]

    Survey on Large Language Model-Enhanced Reinforcement Learning: Concept, Taxonomy, and Methods,

    Y . Cao, H. Zhao, et al. , “Survey on Large Language Model-Enhanced Reinforcement Learning: Concept, Taxonomy, and Methods,” IEEE Trans. on NN and Learning Syst. , pp. 1–21, 2024

    work page 2024

  7. [7]

    A 2.1TFLOPS/W Mobile Deep RL Accelerator with Transposable PE Array and Experience Compression,

    C. Kim, S. Kang, et al., “A 2.1TFLOPS/W Mobile Deep RL Accelerator with Transposable PE Array and Experience Compression,” in ISSCC, pp. 136–138, 2019

    work page 2019

  8. [8]

    Explainable Reinforce- ment Learning: A Survey and Comparative Review,

    S. Milani, N. Topin, M. Veloso, and F. Fang, “Explainable Reinforce- ment Learning: A Survey and Comparative Review,” ACM Comput. Surv., vol. 56, Apr. 2024

    work page 2024

Show all 34 references
  1. [9]

    Efficient and scalable reinforcement learning for large-scale network control,

    C. Ma, A. Li, Y . Du, H. Dong, and Y . Yang, “Efficient and scalable reinforcement learning for large-scale network control,” Nature Machine Intelligence, vol. 6, no. 9, pp. 1006–1020, 2024

    2024

  2. [10]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,

    D. Guo, D. Yang, et al., “Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning,” arXiv preprint 2501.12948 , 2025

    2025 arXiv

  3. [11]

    Flex-PE: Flexible and SIMD Multi-Precision Processing Element for AI Workloads,

    M. Lokhande, G. Raut, and S. K. Vishvakarma, “Flex-PE: Flexible and SIMD Multi-Precision Processing Element for AI Workloads,” IEEE Trans. VLSI Syst. , 2025

    2025

  4. [12]

    Flex-SFU: Activation Function Acceleration with Non-Uniform Piecewise Approximation,

    R. Andri, E. Reggiani, and L. Cavigelli, “Flex-SFU: Activation Function Acceleration with Non-Uniform Piecewise Approximation,”IEEE Trans. Comput.-Aided Design Integr. Circuits Syst. , pp. 1–1, 2025

    2025

  5. [13]

    LPRE: Logarithmic Posit-enabled Reconfigurable edge-AI Engine,

    O. Kokane, M. Lokhande, G. Raut, A. Teman, and S. K. Vishvakarma, “LPRE: Logarithmic Posit-enabled Reconfigurable edge-AI Engine,” ISCAS, 2025

    2025

  6. [14]

    A Reconfigurable Processing Element for Multiple- Precision Floating/Fixed-Point HPC,

    B. Li, K. Li, et al., “A Reconfigurable Processing Element for Multiple- Precision Floating/Fixed-Point HPC,” IEEE Trans. Circuits Syst. II , vol. 71, no. 3, pp. 1401–1405, 2024

    2024

  7. [15]

    A Configurable Floating-Point Multiple-Precision Processing Element for HPC and AI Converged Computing,

    W. Mao, K. Li, et al., “A Configurable Floating-Point Multiple-Precision Processing Element for HPC and AI Converged Computing,” IEEE Trans. VLSI Syst. , vol. 30, no. 2, pp. 213–226, 2022

    2022

  8. [16]

    High-Performance Accurate and Approximate Multipliers for FPGA-Based Hardware Ac- celerators,

    S. Ullah, S. Rehman, M. Shafique, and A. Kumar, “High-Performance Accurate and Approximate Multipliers for FPGA-Based Hardware Ac- celerators,” IEEE Trans. Comput.-Aided Design Integr. Circuits Syst. , vol. 41, no. 2, pp. 211–224, 2022

    2022

  9. [17]

    QuantMAC: Enhancing Hardware Perfor- mance in DNNs With Quantize Enabled Multiply-Accumulate Unit,

    N. Ashar, G. Raut, et al. , “QuantMAC: Enhancing Hardware Perfor- mance in DNNs With Quantize Enabled Multiply-Accumulate Unit,” IEEE Access, vol. 12, pp. 43600–43614, 2024

    2024

  10. [18]

    Unified Posit/IEEE-754 Vector MAC Unit for Transprecision Computing,

    L. Crespo, P. Tom ´as, N. Roma, and N. Neves, “Unified Posit/IEEE-754 Vector MAC Unit for Transprecision Computing,” IEEE Trans. Circuits Syst. II, vol. 69, no. 5, pp. 2478–2482, 2022

    2022

  11. [19]

    A Low-Cost Floating-Point FMA Unit Supporting Package Operations for HPC-AI Applications,

    H. Tan, J. Zhang, et al. , “A Low-Cost Floating-Point FMA Unit Supporting Package Operations for HPC-AI Applications,” IEEE Trans. Circuits Syst. II , vol. 71, no. 7, pp. 3488–3492, 2024

    2024

  12. [20]

    A Low-Cost Floating-Point Dot-Product-Dual- Accumulate Architecture for HPC-Enabled AI,

    H. Tan, L. Huang, et al., “A Low-Cost Floating-Point Dot-Product-Dual- Accumulate Architecture for HPC-Enabled AI,” IEEE Trans. Comput.- Aided Design Integr. Circuits Syst. , vol. 43, no. 2, pp. 681–693, 2024

    2024

  13. [21]

    A Vector Systolic Accelerator for Multi- Precision Floating-Point High-Performance Computing,

    K. Li, W. Mao, et al. , “A Vector Systolic Accelerator for Multi- Precision Floating-Point High-Performance Computing,” IEEE Trans. Circuits Syst. II , vol. 69, no. 10, pp. 4123–4127, 2022

    2022

  14. [22]

    Multiple-Mode- Supporting Floating-Point FMA Unit for Deep Learning Processors,

    H. Tan, G. Tong, L. Huang, L. Xiao, and N. Xiao, “Multiple-Mode- Supporting Floating-Point FMA Unit for Deep Learning Processors,” IEEE Trans. VLSI Syst. , vol. 31, no. 2, pp. 253–266, 2023

    2023

  15. [23]

    A Two-Stage Operand Trimming Approximate Logarithmic Multiplier,

    R. Pilipovi ´c, P. Buli´c, and U. Lotri ˇc, “A Two-Stage Operand Trimming Approximate Logarithmic Multiplier,” IEEE Trans. Circuits Syst. I , vol. 68, no. 6, pp. 2535–2545, 2022

    2022

  16. [24]

    An Empirical Approach to Enhance Performance for Scalable CORDIC-Based Deep Neural Networks,

    G. Raut, S. Karkun, and S. K. Vishvakarma, “An Empirical Approach to Enhance Performance for Scalable CORDIC-Based Deep Neural Networks,” ACM Trans. Recon. Technol. Syst. , vol. 16, June 2023

    2023

  17. [25]

    Efficient CORDIC-Based Activation Functions for RNN Acceleration on FPGAs,

    W. Shen, J. Jiang, M. Li, and S. Liu, “Efficient CORDIC-Based Activation Functions for RNN Acceleration on FPGAs,” IEEE Trans. on Artificial Intelligence , vol. 6, no. 1, pp. 199–210, 2025

    2025

  18. [26]

    Approximate Softmax Functions for Energy-Efficient Deep Neural Networks,

    K. Chen, Y . Gao, H. Waris, W. Liu, and F. Lombardi, “Approximate Softmax Functions for Energy-Efficient Deep Neural Networks,” IEEE Trans. VLSI Syst. , vol. 31, no. 1, pp. 4–16, 2023

    2023

  19. [27]

    A Unified Parallel CORDIC- Based Hardware Architecture for LSTM Network Acceleration,

    N. A. Mohamed and J. R. Cavallaro, “A Unified Parallel CORDIC- Based Hardware Architecture for LSTM Network Acceleration,” IEEE Trans. on Computers, vol. 72, no. 10, pp. 2752–2766, 2023

    2023

  20. [28]

    Synergy: An HW/SW Framework for High Throughput CNNs on Embedded Heterogeneous SoC,

    G. Zhong, A. Dubey, C. Tan, and T. Mitra, “Synergy: An HW/SW Framework for High Throughput CNNs on Embedded Heterogeneous SoC,” ACM Trans. Embed. Comput. Syst. , vol. 18, Mar. 2019

    2019

  21. [29]

    Real-Time SSDLite Object Detection on FPGA,

    S. Kim, S. Na, B. Y . Kong, J. Choi, and I.-C. Park, “Real-Time SSDLite Object Detection on FPGA,” IEEE Trans. VLSI Syst. , vol. 29, no. 6, pp. 1192–1205, 2022

    2022

  22. [30]

    ShortcutFusion: From Tensorflow to FPGA-Based Accelerator With a Reuse-Aware Memory Allocation for Shortcut Data,

    D. T. Nguyen, H. Je, T. N. Nguyen, S. Ryu, K. Lee, and H.-J. Lee, “ShortcutFusion: From Tensorflow to FPGA-Based Accelerator With a Reuse-Aware Memory Allocation for Shortcut Data,” IEEE Trans. Circuits Syst. I , vol. 69, no. 6, pp. 2477–2489, 2022

    2022

  23. [31]

    A High-Throughput Full-Dataflow Mo- bileNetv2 Accelerator on Edge FPGA,

    W. Jiang, H. Yu, and Y . Ha, “A High-Throughput Full-Dataflow Mo- bileNetv2 Accelerator on Edge FPGA,” IEEE Trans. Comput.-Aided Design Integr. Circuits Syst., vol. 42, no. 5, pp. 1532–1545, 2023

    2023

  24. [32]

    A Real-Time Object Detection Processor With xnor-Based Variable-Precision Computing Unit,

    W. Lee, K. Kim, W. Ahn, J. Kim, and D. Jeon, “A Real-Time Object Detection Processor With xnor-Based Variable-Precision Computing Unit,” IEEE Trans. VLSI Syst. , vol. 31, no. 6, pp. 749–761, 2023

    2023

  25. [33]

    Edge-Side Fine-Grained Sparse CNN Accelerator With Efficient Dynamic Pruning Scheme,

    B. Wu, T. Yu, K. Chen, and W. Liu, “Edge-Side Fine-Grained Sparse CNN Accelerator With Efficient Dynamic Pruning Scheme,” IEEE Trans. Circuits Syst. I , vol. 71, no. 3, pp. 1285–1298, 2024

    2024

  26. [34]

    Low Latency Hybrid CORDIC Algorithm,

    R. Shukla and K. C. Ray, “Low Latency Hybrid CORDIC Algorithm,” IEEE Trans. on Computers , vol. 63, no. 12, pp. 3066–3078, 2014

    2014

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.