Pith. sign in

REVIEW 2 major objections 5 minor 52 references

OASIS: Optimized Lightweight Autoencoder System for Distributed In-Sensor computing

T0 review · 2 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A dual-branch autoencoder with entropy-regularized training compresses sensor output up to 11,985x, cutting total system energy by 2-4.5x while keeping accuracy near baseline.

desk verdict The in-sensor compression idea and task results are plausible, but the headline 2–4.5× energy savings rest on an arithmetic error that overstates baseline MIPI energy by ~6.6×. read the letter →

arxiv 2505.02256 v1 pith:MKWIEKOQ submitted 2025-05-04 eess.IV

Chengwei Zhou , Sreetama Sarkar , Yuming Li , Arnab Sanyal , Gourav Datta This is my paper · ORCID
classification eess.IV
keywords in-sensorcomputingautoencoderactivationcompressionentropylossHuffmancodingquantization-awaretrainingenergy-efficientvisionAR/VRworkloads
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

This paper argues that the data bottleneck that makes in-sensor computing impractical can be removed by training a tiny autoencoder to emit a highly compressed code for the vision task itself. Its dual-branch design keeps a lightweight encoder on the sensor-side logic chip, uses a decoder only during training to preserve visual structure, and sends the learned code off-chip after 4-bit quantization and Huffman coding. The reported result is a bandwidth reduction of up to $11{,}985\times$ relative to the input image, a $2{-}4.5\times$ drop in total system energy under a hardware-backed energy model, and accuracy near baseline on person-detection, hand-tracking, and eye-tracking workloads. If this holds, in-sensor processors no longer need to run deep layers; they only need to compute a compact feature code for a cheap off-chip processor to decode.

What carries the argument

The load-bearing mechanism is the dual-branch autoencoder with a triple training objective. The encoder $\mathcal{E}$ lives on the sensor-side logic chip; the decoder $\mathcal{D}$ uses an expand-contract-expand schedule to reconstruct the input during training, forcing the low-dimensional latent to preserve visual information; the off-chip task network consumes the de-quantized latent. The objective $L = \mathrm{MSE}(X,\hat{X}) + \beta L_{\mathrm{entropy}} + \gamma L_{\mathrm{task}}$ ties these branches together, where $L_{\mathrm{entropy}}$ is a hinge loss on the empirical entropy of the quantized encoder output above $H_{\mathrm{ref}}=0.7$ bits, computed by histogram analysis. Quantization-aware training with a straight-through estimator makes the 4-bit code robust, and the skewed symbol distribution it creates is what makes Huffman coding drop the effective bit width to 1.57 bits. The energy model then converts byte counts into joules: APS read-out and ADC, TSV transfer at 6.25 pJ/byte, MIPI transfer at 100 pJ/byte, MAC compute at 5 fJ, and SRAM reads at 1.84 pJ per 8-bit weight.

What would settle it

Run the trained quantized encoder on a real 7nm or 22nm sensor-plus-logic-chip stack and measure (a) the actual per-byte MIPI energy and (b) the actual Huffman-coded bit rate over a large test set; if per-byte energy is well below 100 pJ, or the average code length exceeds about 1.57 bits per symbol, the total-energy model and the $2{-}4.5\times$ savings will not be reproduced.

Watch

Extended reading notes

Core claim

The central claim is that a vision encoder can be made tiny enough to fit next to a CMOS image sensor while still producing a representation that is simultaneously good for the downstream task and extremely cheap to transmit. The paper shows this by training encoder and task network end-to-end with a triple loss: a reconstruction loss through an expand-contract-expand decoder (active only in training), an entropy loss that caps the empirical entropy of the quantized code at $H_{\mathrm{ref}}=0.7$ bits, and the task loss. With this training, a 4-bit quantized encoder output with spatial size $4\times4$ and 4 channels compresses a $224\times224\times3$ image by $2{,}352\times$, and Huffman coding on the skewed symbol distribution brings the effective bit width from 4 to 1.57 bits, for a total $11{,}985\times$ bandwidth reduction. At that setting the VWW classifier stays within 1% of baseline accuracy; hand-tracking 2D AUC is $0.880$ vs. $0.884$ baseline and 3D AUC $0.969$ vs. $0.973$; eye-tracking mIoU is $0.974$ vs. $0.988$. The total per-frame energy falls to roughly half to under a quarter of a baseline that sends the full image over the MIPI interface, because MIPI transfer at 100 pJ/byte dominated the baseline's 169.9$\mu$J per frame.

Load-bearing premise

The load-bearing premise is that sending one byte over the sensor's off-chip MIPI interface costs about 100 pJ, so shrinking the byte count is what saves energy; if the real interface energy is much lower, or if the Huffman stream does not actually reach the 1.57-bit average in deployment, the claimed $2{-}4.5\times$ total-energy reduction shrinks.

Editorial extensions

If this is right

  • In-sensor processors can be built around a very small encoder rather than a full network, because the decoder needed for training does not run at inference time.
  • AR/VR workloads such as eye tracking and hand tracking can move most of their data movement off the MIPI bottleneck; the paper reports 22.7 TOPS/W at near-baseline accuracy.
  • The total compression factor is a product of spatial reduction, bit-width reduction, and entropy coding, so tasks whose learned latent distributions are similarly skewed will see comparable factors.
  • Because the encoder output fits in on-chip SRAM, off-chip DRAM traffic can be avoided, making system energy scale with the transmitted code size rather than input resolution.
  • The same architecture keeps near-baseline accuracy on high-resolution inputs while compressing by four orders of magnitude, moving beyond the small benchmark images used by many earlier in-sensor designs.

Reading between the lines

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

  • The low entropy that enables 1.57 bits per symbol is actively manufactured by the entropy loss during training; it is a property of the learned code, not of natural images, so the compression ratio should be expected to change with $H_{\mathrm{ref}}$ and the task.
  • A natural deployment test is to fix the Huffman table and measure its true on-wire overhead and worst-case code length; the paper reports the average 1.57 bits/symbol but not the table-size or tail-length cost.
  • The same dual-branch recipe could transfer to other high-bandwidth modalities such as LiDAR point clouds, radar, or event streams, but the encoder, entropy target, and Huffman table would need to be retrained per modality.
  • Because reconstruction only shapes the latent during training, the off-chip task network could in principle be swapped without retraining the in-sensor encoder, although the paper does not test that transfer scenario.
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

2 major / 5 minor

Summary. The manuscript proposes OASIS, a dual-branch autoencoder architecture for in-sensor computing. A lightweight encoder runs on a logic chip stacked with the image sensor; a decoder is used only during training; and a task-specific network runs off-chip. The encoder is trained with a combination of reconstruction, entropy, and task losses, and its 4-bit quantized output is Huffman-coded before transmission over the sensor interface. The paper reports up to 11985x bandwidth reduction on Visual Wake Words and claims 2-4.5x system-level energy savings across VWW classification, hand tracking, and eye tracking, while keeping accuracy close to baselines. The energy results are obtained from a semi-analytical model with per-component terms for APS, TSV, MIPI, encoder, and backend processor energy.

Significance. If the energy model were correct, the paper would make a useful contribution: the compression arithmetic is transparent (2352x from dimensionality, 2x from 8-bit to 4-bit quantization, and about 2.55x from Huffman coding), the training objective is clearly specified, and the evaluation covers three realistic vision tasks. The dual-branch scheme with a training-time decoder and a deployment-time encoder is a sensible way to learn task-relevant compression. The strongest aspects are the explicit, reproducible compression pipeline and the breadth of the task evaluation. The main weakness is the system-level energy validation, which currently contains an arithmetic inconsistency that directly affects the central 2-4.5x energy claim.

major comments (2)
  1. [§IV-C2, §V-B1, Eq. (3)] Eq. (3) and §IV-C2 set E_Byte,inf = 100 pJ/byte. For the VWW baseline of a 224x224x3 8-bit image, the MIPI transfer cost is therefore 224*224*3*100 pJ = 15.05 uJ, not the 99.5 uJ stated in §V-B1. The 99.5 uJ figure corresponds to about 661 pJ/byte. The paper uses this 99.5 uJ figure as the dominant term of the 169.9 uJ baseline and derives the 2-4.5x energy savings and Fig. 5(a) from it, so the central energy claim is not supported by the paper's own equations. Please correct the per-byte energy, the byte count, or the baseline MIPI energy, and re-derive all system-level energy results with a full component-by-component breakdown.
  2. [§IV-C2, §V-B1] The energy conclusion is also highly sensitive to the assumed MIPI per-byte energy and to the effective Huffman width of 1.57 bits. The former rests on a single reference [37], and the latter is an average that depends on the exact quantized-output histogram produced under Href = 0.7. Please add a sensitivity analysis, varying E_Byte,inf over a plausible range and reporting the measured Huffman bit width at deployment, and show whether the 2-4.5x conclusion survives. Without such an analysis, the headline energy reduction is a consequence of one unvalidated per-byte assumption.
minor comments (5)
  1. [Abstract, §I] The phrase 'four-order-of-magnitude reduction in output activation dimensionality' is imprecise; the VWW raw dimensionality reduction is 2352x (about 3.4 orders), and the 11985x figure is a bandwidth reduction after quantization and Huffman coding. Please reword to distinguish dimensionality reduction from bit-rate reduction.
  2. [§V-B1] The text reports 'MIPI energy is 99.5 uJ' without showing the multiplication; please include the explicit byte count and per-byte energy in the main text so the calculation can be checked directly.
  3. [§IV-C3] The extrapolation of Huffman encoding/decoding energies from the 45nm implementation [40] to 7nm is stated without a scaling model; please describe the extrapolation or provide a range of values.
  4. [Table II] The TOPS/W comparison mixes different tasks, input resolutions, and technology nodes; please add a footnote stating that the numbers are not directly comparable across rows.
  5. [§II-B, Eq. (2)] The term E_{Zquant~P}[-log2 P(Zquant)] should be identified as the entropy of the quantized output, and the histogram-based construction of P should be described in the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the compression ratio is arithmetic from the chosen architecture and measured Huffman bit-width; the energy conclusion is computed from an explicit semi-analytical model. The 99.5 µJ MIPI baseline is internally inconsistent with the stated 100 pJ/byte, but that is a correctness issue, not circularity.

full rationale

The derivation chain is self-contained. The 11985× bandwidth reduction is a byte-count ratio: 224×224×3 8-bit input bytes versus a 4×4×4 encoder output quantized to 4 bits and Huffman-coded at a measured 1.57-bit average width. The entropy loss in Eq. (2) does explicitly encourage low entropy with Href=0.7, but the realized 1.57-bit width is a measured statistic on evaluation distributions (Fig. 1), lies above Href, and is not used as an input that is then reported as an output. Reporting the achieved bit-rate of a trained compression model is evaluation, not circular prediction. Accuracy claims are external benchmark comparisons (VWW, STB, OpenEDS), and the 2–4.5× energy savings are computed from Eq. (3), a sum of separately sourced APS, TSV, MIPI, encoder, and back-end energies. Self-citations to [32], [33] for the energy-model form are not load-bearing because the model is fully stated in the paper. One non-circular correctness flag: Section V-B1 says the VWW baseline MIPI energy is 99.5 µJ, but Eq. (3) states EByte,inf=100 pJ/byte, which gives 224·224·3·100 pJ = 15.05 µJ; the 99.5 µJ figure would require about 661 pJ/byte. This arithmetic inconsistency affects the claimed energy-reduction factor, but it is not a circular-derivation issue.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central energy claim rests on external per-byte energy constants, an unreleased in-house APS simulation, and an extrapolated Huffman energy. The compression factor depends on hand-chosen encoder output dimensions and loss weights. No new physical entities are introduced, but several numeric inputs are taken from prior work or ad hoc extrapolation without independent verification in this paper.

free parameters (4)
  • Encoder output dimension (d channels, s x s spatial) = VWW: 4 channels, 4x4 spatial; hand/eye: d and s varied across configurations
    Chosen by hand per task; this directly sets the bandwidth reduction factor, e.g., 2352x before quantization for the VWW configuration.
  • Entropy target Href = 0.7
    Used in Eq. (2) to control the rate-distortion trade-off and to force low entropy output, which in turn produces the claimed Huffman effective bit width of 1.57 bits.
  • Loss weights beta and gamma = beta=2, gamma=2 for VWW; beta=2, gamma=4 for hand and eye tracking
    Hyperparameters that balance reconstruction, entropy, and task accuracy; no sensitivity analysis is provided, and the results depend on these choices.
  • Encoder output bit precision n = 4 bits
    Quantization level for the compressed stream; the reported compression factor includes a 2x gain from using 4 bits instead of 8-bit input pixels.
assumptions (5)
  • domain assumption MIPI interface energy is 100 pJ per byte [37]
    Used in Eq. (3) and Section IV-C2; MIPI transfer dominates baseline energy (99.5 of 169.9 uJ per frame), so a lower real value would shrink the claimed 2 to 4.5x savings.
  • domain assumption APS readout plus ADC energy is 63.6 pJ per pixel from in-house 22nm GF FDSOI simulations
    Section IV-B states this value but does not describe the simulation; it cannot be checked from the paper.
  • domain assumption TSV energy is 6.25 pJ per byte [35]
    Section IV-C1 uses this for the added internal interconnect cost; it is small compared with MIPI and does not drive the main result.
  • domain assumption MAC energy of 5 fJ and SRAM read energy of 1.84 pJ per 8-bit weight from prior 7nm data [38], [39]
    Section IV-D uses these constants to estimate encoder and backend compute and memory energy; they come from prior hardware papers, not from measurements in this work.
  • ad hoc to paper Huffman encoding and decoding energies of 0.96 and 1.15 pJ per byte in 7nm, extrapolated from a 45nm implementation [40]
    Section IV-D3 extrapolates from a 45nm design to 7nm using unspecified process scaling trends; the extrapolation is not shown and the value is unverified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OASIS: Optimized Lightweight Autoencoder System for Distributed In-Sensor computing." pith.science (2026). https://pith.science/paper/MKWIEKOQ

@misc{pith2026250502256,
  author       = {Pith},
  title        = {Pith review of: OASIS: Optimized Lightweight Autoencoder System for Distributed In-Sensor computing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MKWIEKOQ}},
  note         = {Machine review of arXiv:2505.02256}
}
abstract

In-sensor computing, which integrates computation directly within the sensor, has emerged as a promising paradigm for machine vision applications such as AR/VR and smart home systems. By processing data on-chip before transmission, it alleviates the bandwidth bottleneck caused by high-resolution, high-frame-rate image transmission, particularly in video applications. We envision a system architecture that integrates a CMOS image sensor (CIS) with a logic chip via advanced packaging, where the logic chip processes early-stage deep neural network (DNN) layers. However, its limited compute and memory make deploying advanced DNNs challenging. A simple solution is to split the model, executing the first part on the logic chip and the rest off-chip. However, modern DNNs require multiple layers before dimensionality reduction, limiting their ability to achieve the primary goal of in-sensor computing: minimizing data bandwidth. To address this, we propose a dual-branch autoencoder-based vision architecture that deploys a lightweight encoder on the logic chip while the task-specific network runs off-chip. The encoder is trained using a triple loss function: (1) task-specific loss to optimize accuracy, (2) entropy loss to enforce compact and compressible representations, and (3) reconstruction loss (mean-square error) to preserve essential visual information. This design enables a four-order-of-magnitude reduction in output activation dimensionality compared to input images, resulting in a $2{-}4.5\times$ decrease in energy consumption, as validated by our hardware-backed semi-analytical energy models. We evaluate our approach on CNN and ViT-based models across applications in smart home and augmented reality domains, achieving state-of-the-art accuracy with energy efficiency of up to 22.7 TOPS/W.

Figures

Figures reproduced from arXiv: 2505.02256 by the authors.

Figure 1
Figure 1. Probability distribution of the 4-bit quantized encoder output for input [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the dual-branch autoencoder-based in-sensor computing architecture. The system integrates an [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Accuracy vs. Representation Dimension comparison on VWW [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: (a) Normalized energy consumption comparison between our approach [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 48 canonical work pages

  1. [37]

    A 3.0 Gsymbol/s/lane MIPI C-PHY receiver with adaptive level-dependent equalizer for mobile CMOS image sensor

    Seokwon Choi et al. A 3.0 Gsymbol/s/lane MIPI C-PHY receiver with adaptive level-dependent equalizer for mobile CMOS image sensor. Sensors, 21(15), 2021

    work page 2021

  2. [1]

    Deep learning-based computer vision for surveillance in its: Evaluation of state-of-the-art methods

    Jiyang Xie et al. Deep learning-based computer vision for surveillance in its: Evaluation of state-of-the-art methods. IEEE Transactions on Vehicular Technology, 70(4):3027–3042, 2021

    work page 2021

  3. [2]

    How computer vision can facilitate flood management: A systematic review

    Umair Iqbal et al. How computer vision can facilitate flood management: A systematic review. International Journal of Disaster Risk Reduction , 53:102030, 2021

    work page 2021

  4. [3]

    E.R. Fossum. CMOS image sensors: electronic camera-on-a-chip. IEEE Transactions on Electron Devices , 44(10):1689–1698, 1997

    work page 1997

  5. [4]

    Near-sensor distributed dnn processing for aug- mented and virtual reality

    Reid Pinkham et al. Near-sensor distributed dnn processing for aug- mented and virtual reality. IEEE Journal on Emerging and Selected Topics in Circuits and Systems , 11(4):663–676, 2021

    work page 2021

  6. [5]

    https://www.sony.com/en/SonyInfo/News/Press/ 202005/20-037E/, 2020

    Sony to Release World’s First Intelligent Vision Sensors with AI Pro- cessing Functionality. https://www.sony.com/en/SonyInfo/News/Press/ 202005/20-037E/, 2020. Accessed: 12-01-2022

    work page 2020

  7. [6]

    Processing near sensor architecture in mixed- signal domain with cmos image sensor of convolutional-kernel-readout method

    Zhe Chen et al. Processing near sensor architecture in mixed- signal domain with cmos image sensor of convolutional-kernel-readout method. IEEE Transactions on Circuits and Systems I: Regular Papers , 67(2):389–400, 2020

    work page 2020

  8. [7]

    Ultrafast machine vision with 2D material neural network image sensors

    Lukas Mennel et al. Ultrafast machine vision with 2D material neural network image sensors. Nature, 579:62–66, 2020

    work page 2020

Show all 52 references
  1. [8]

    Fully embedding fast convolutional networks on pixel processor arrays

    Laurie Bose et al. Fully embedding fast convolutional networks on pixel processor arrays. In Computer Vision - ECCV 2020 - 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceedings, Part XXIX, volume 12374, pages 488–503. Springer, 2020

    2020

  2. [9]

    A reconfigurable convolution-in-pixel CMOS image sensor architecture

    Ruibing Song et al. A reconfigurable convolution-in-pixel CMOS image sensor architecture. IEEE Transactions on Circuits and Systems for Video Technology, 2022

    2022

  3. [10]

    Integrated pixel and two-terminal non- volatile memory cell and an array of cells for deep in-sensor, in-memory computing, December 7 2021

    Akhilesh Jaiswal and Ajey Jacob. Integrated pixel and two-terminal non- volatile memory cell and an array of cells for deep in-sensor, in-memory computing, December 7 2021. US Patent 11,195,580

    2021

  4. [11]

    Pisa: A non-volatile processing-in-sensor accel- erator for imaging systems

    Shaahin Angizi et al. Pisa: A non-volatile processing-in-sensor accel- erator for imaging systems. IEEE Transactions on Emerging Topics in Computing, 11(4):962–972, 2023

    2023

  5. [12]

    P2M-DeTrack: Processing-in-pixel-in-memory for energy-efficient and real-time multi-object detection and tracking

    Gourav Datta et al. P2M-DeTrack: Processing-in-pixel-in-memory for energy-efficient and real-time multi-object detection and tracking. In 2022 IFIP/IEEE 30th International Conference on Very Large Scale Integration (VLSI-SoC), pages 1–6. IEEE, 2022

    2022

  6. [13]

    P2M: A processing-in-pixel-in-memory paradigm for resource-constrained TinyML applications

    Gourav Datta et al. P2M: A processing-in-pixel-in-memory paradigm for resource-constrained TinyML applications. Scientific Reports , 12(1):14396, 2022

    2022

  7. [14]

    Jacob, Pedram Khalili Amiri, Peter A

    Md Abdullah-Al Kaiser, Gourav Datta, Jordan Athas, Christian Duffee, Ajey P. Jacob, Pedram Khalili Amiri, Peter A. Beerel, and Akhilesh R. Jaiswal. V oltage-controlled magnetic tunnel junction based adc-less global shutter processing-in-pixel for extreme-edge intelligence. arX...

    arXiv

  8. [15]

    Beerel, and Akhilesh R

    Md Abdullah-Al Kaiser, Gourav Datta, Peter A. Beerel, and Akhilesh R. Jaiswal. Toward high-accuracy, programmable extreme-edge intelligence for neuromorphic vision sensors utilizing magnetic domain wall motion- based mtj. In Proceedings of the 61st ACM/IEEE Design Automation C...

    2024

  9. [16]

    Distributed on-sensor compute system for AR/VR devices: A semi-analytical simulation framework for power estimation

    Jorge Gomez et al. Distributed on-sensor compute system for AR/VR devices: A semi-analytical simulation framework for power estimation. In Proceedings of the tinyML Research Symposium , page 6, 2022

    2022

  10. [17]

    Estimating power, performance, and area for on-sensor deployment of ar/vr workloads using an analytical frame- work

    Jorge Gomez et al. Estimating power, performance, and area for on-sensor deployment of ar/vr workloads using an analytical frame- work. ACM Journal on Emerging Technologies in Computing Systems , 19(4):1–20, 2023

    2023

  11. [18]

    Splitnets: Designing neural architectures for efficient distributed computing on head-mounted systems

    Xin Dong et al. Splitnets: Designing neural architectures for efficient distributed computing on head-mounted systems. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12559–12569, June 2022

    2022

  12. [19]

    Gregory K. Wallace. The jpeg still picture compression standard. Communications of the ACM , 34(4):30–44, 1991

    1991

  13. [20]

    Torfason et al

    R. Torfason et al. Towards image understanding from deep compression without decoding. In ICLR, 2018

    2018

  14. [21]

    Deep residual learning for image recognition

    Kaiming He et al. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  15. [22]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu et al. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international con- ference on computer vision , pages 10012–10022, 2021

    2021

  16. [23]

    Soft-to-hard vector quantization for end-to-end learning compressible representations

    Eirikur Agustsson et al. Soft-to-hard vector quantization for end-to-end learning compressible representations. Advances in neural information processing systems, 30, 2017

    2017

  17. [24]

    Quantization and training of neural networks for efficient integer-arithmetic-only inference

    Benoit Jacob et al. Quantization and training of neural networks for efficient integer-arithmetic-only inference. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2704–2713, 2018

    2018

  18. [25]

    Estimating or propagating gradients through stochastic neurons for conditional computation

    Yoshua Bengio et al. Estimating or propagating gradients through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432, 2013

    2013 arXiv

  19. [26]

    I-vit: Integer-only quantization for efficient vision transformer inference

    Zhikai Li and Qingyi Gu. I-vit: Integer-only quantization for efficient vision transformer inference. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision , pages 17065–17075, 2023

    2023

  20. [27]

    David A. Huffman. A method for the construction of minimum- redundancy codes. Proceedings of the IRE , 40(9):1098–1101, 1952

    1952

  21. [28]

    Cmos image sensors: Electronic camera-on-a-chip

    Eric R Fossum. Cmos image sensors: Electronic camera-on-a-chip. IEEE Transactions on Electron Devices , 44(10):1689–1698, 1997

    1997

  22. [29]

    Kawahito

    S. Kawahito. Survey of column-parallel ADC architectures for CMOS image sensors. IEICE Transactions on Electronics, 101(8):53–61, 2018

    2018

  23. [30]

    Hsu et al

    T.-H. Hsu et al. 0.8V CNN Image Sensor with mixed-signal processing- in-sensor architecture. IEEE Journal of Solid-State Circuits, 58(2):1064– 1072, 2023

    2023

  24. [31]

    Venus: A versatile deep neural network accelerator architecture design for multiple applications

    Jiaqi Yang et al. Venus: A versatile deep neural network accelerator architecture design for multiple applications. In Proceedings of the 60th ACM/IEEE Design Automation Conference (DAC) , pages 1–6, 2023

    2023

  25. [32]

    Jacob, Peter A

    Md Abdullah-Al Kaiser, Gourav Datta, Sreetama Sarkar, Souvik Kundu, Zihan Yin, Manas Garg, Ajey P. Jacob, Peter A. Beerel, and Akhilesh R. Jaiswal. Technology-circuit-algorithm tri-design for processing-in-pixel- in-memory (p2m). In Proceedings of the Great Lakes Symposium on ...

    2023

  26. [33]

    Jacob, Akhilesh R

    Gourav Datta, Zeyu Liu, Md Abdullah-Al Kaiser, Souvik Kundu, Joe Mathai, Zihan Yin, Ajey P. Jacob, Akhilesh R. Jaiswal, and Peter A. Beerel. In-sensor & neuromorphic computing are all you need for energy efficient computer vision. In ICASSP 2023 - 2023 IEEE International Confe...

    2023

  27. [34]

    Liu and Y

    X. Liu and Y . Chen. Power consumption analysis of CMOS image sensors for energy-efficient vision systems. IEEE Sensors Journal , 20(9):1054–1062, 2020

    2020

  28. [35]

    Vivet and A

    P. Vivet and A. Fayed. Through-silicon via (TSV) technologies for low- power interconnects. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 28(11):2501–2513, 2020

    2020

  29. [36]

    Choi et al

    K. Choi et al. High-speed MIPI interfaces for embedded vision systems. IEEE Transactions on Consumer Electronics , 67(3):450–458, 2021

    2021

  30. [38]

    M. E. Sinangil et al. A 7-nm compute-in-memory SRAM macro supporting multi-bit input, weight and output and achieving 351 tops/w. IEEE Journal of Solid-State Circuits , 55(12):3488–3501, 2020

    2020

  31. [39]

    Robust 7-nm SRAM design on a predictive PDK

    Various Authors. Robust 7-nm SRAM design on a predictive PDK. ResearchGate, 2017

    2017

  32. [40]

    Low power text compression for huffman coding using altera FPGA

    Salih Bayar et al. Low power text compression for huffman coding using altera FPGA. In 2018 International Conference on Artificial Intelligence and Data Processing (IDAP) , pages 1–6. IEEE, 2018

    2018

  33. [41]

    Visual wake words dataset

    Aakanksha Chowdhery et al. Visual wake words dataset. arXiv preprint arXiv:1906.05721, 2019

    1906 arXiv

  34. [42]

    Real-time gaze tracking with event-driven eye segmenta- tion

    Yu Feng et al. Real-time gaze tracking with event-driven eye segmenta- tion. In 2022 IEEE Conference on Virtual Reality and 3D User Interfaces (VR), pages 399–408. IEEE, 2022

    2022

  35. [43]

    A hand pose tracking benchmark from stereo matching

    Jiawei Zhang et al. A hand pose tracking benchmark from stereo matching. In 2017 IEEE International Conference on Image Processing (ICIP), pages 982–986. IEEE, 2017

    2017

  36. [44]

    Megatrack: monochrome egocentric articulated hand-tracking for virtual reality

    Shangchen Han et al. Megatrack: monochrome egocentric articulated hand-tracking for virtual reality. ACM Transactions on Graphics (ToG), 39(4):87–1, 2020

    2020

  37. [45]

    Openeds2020: Open eyes dataset

    Cristina Palmero et al. Openeds2020: Open eyes dataset. arXiv preprint arXiv:2005.03876, 2020

    2005 arXiv

  38. [46]

    Energy-efficient & real-time computer vision with intelligent skipping via reconfigurable CMOS image sensors

    Md Abdullah-Al Kaiser et al. Energy-efficient & real-time computer vision with intelligent skipping via reconfigurable CMOS image sensors. arXiv preprint arXiv:2409.17341 , 2024

    2024 arXiv

  39. [47]

    Utilizing direct photocurrent computation and 2d kernel scheduling to improve in-sensor-processing efficiency

    Han Xu et al. Utilizing direct photocurrent computation and 2d kernel scheduling to improve in-sensor-processing efficiency. In 2020 57th ACM/IEEE Design Automation Conference (DAC) . IEEE, 2020

    2020

  40. [48]

    A reconfigurable convolution-in-pixel CMOS image sensor architecture

    Ruibing Song et al. A reconfigurable convolution-in-pixel CMOS image sensor architecture. arXiv preprint arXiv:2101.03308 , 2021

    2021 arXiv

  41. [49]

    Xu and et al

    H. Xu and et al. Senputing: Always-on vision sensor with sensing- computing fusion. IEEE Transactions on Circuits and Systems I , 69(7):1047–1051, 2022

    2022

  42. [50]

    MR-PIPA: An integrated multilevel RRAM (HfOx)-Based processing-in-pixel accelerator

    Minhaz Abedin et al. MR-PIPA: An integrated multilevel RRAM (HfOx)-Based processing-in-pixel accelerator. IEEE Journal on Ex- ploratory Solid-State Computational Devices and Circuits , 8(2):59–67, 2022

    2022

  43. [51]

    PiPSim: A behavior-level modeling tool for CNN processing-in-pixel accelerators

    Arman Roohi et al. PiPSim: A behavior-level modeling tool for CNN processing-in-pixel accelerators. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems , 2023

    2023

  44. [52]

    Pascal Vivet et al. A 220GOPS 96-Core processor with 6 chiplets 3D- Stacked on an active interposer offering 0.6ns/mm latency, 3Tb/s/mm 2 Inter-Chiplet Interconnects and 156mW/mm 2 at 82%-Peak-Efficiency DC-DC Converters. In 2020 IEEE International Solid-State Circuits Confere...

    2020

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.