Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

FlexiBit: Fully Flexible Precision Bit-parallel Accelerator Architecture for Arbitrary Mixed Precision AI

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read FlexiBit is an accelerator architecture that processes any FP/INT precision and format in a bit-parallel manner, eliminating the padding waste that fixed-width units impose on non-power-of-two formats such as FP6.

desk verdict Genuinely new bit-parallel flexible-precision architecture with real RTL behind it; the headline efficiency numbers rest on modeled baselines that need more scrutiny. read the letter →

arxiv 2411.18065 v2 pith:RCOTM2AM submitted 2024-11-27 cs.AR

Faraz Tahmasebi , Yian Wang , Benji Y.H. Huang , Hyoukjun Kwon This is my paper · ORCID
classification cs.AR
keywords MLacceleratorLargelanguagemodelsQuantizationFlexibleprecisionMixedBit-parallelarchitectureFloating-pointformatsNon-power-of-two
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

FlexiBit is an accelerator architecture whose processing element computes floating-point or integer multiply-accumulate at any bit width and any exponent/mantissa split, in bit-parallel fashion, so non-power-of-two formats like FP6 do not have to be padded into FP8/FP16 containers. The paper argues that this removes a hardware-imposed constraint that has kept quantization algorithms such as FP6-LLM from delivering their algorithmic gains on real silicon. If the architecture works as claimed, it would let LLM inference run FP5/FP6 mixed precision at full compute utilization, yielding 1.66x higher performance per area on GPT-3 in FP6 than a Tensor Core-like unit, 1.62x higher than the flexible-precision BitFusion baseline, and 3.9x higher than bit-serial flexible-precision designs. The claimed area cost of this flexibility is small, roughly 0.5-1% over the fixed-precision and BitFusion baselines.

What carries the argument

The load-bearing object is the Flexible Bit Reduction Tree (FBRT), a fat-tree of switches in which each node can route, shift, concatenate, or add incoming bit-level partial products; it extends the augmented reduction tree from the MAERI design from data-element routing to bit-level routing. FBRT consumes the AND primitives produced by the Primitive Generator and combines them by spatial shift-add, generating multiple mantissa products at once from packed operand registers. It is what lets arbitrary mantissa widths, for example a 3-bit mantissa times a 2-bit mantissa inside FP6/FP5 operations, execute with no idle logic, so the entire architectural claim rests on this tree's reconfigurability. Supporting mechanisms are the segmentable carry-chain exponent adder, the concat-shift tree for exponent alignment, and the crossbar-based bit packing unit.

What would settle it

Build FlexiBit's processing element in RTL at the Cloud-B scale and measure throughput per area for FP6 GEMMs against a Tensor Core-like unit on the same technology; if the measured ratio does not come near the claimed 1.66x, the central efficiency claim fails. A cheaper test: measure compute-unit utilization for a single FP6 mantissa multiply on FBRT; if any mantissa bit lanes idle for arbitrary widths, the 'no underutilization' assertion breaks.

Watch

Extended reading notes

Core claim

The central claim is that bit-level reconfigurability, not new arithmetic, is what unlocks arbitrary mixed precision. FlexiBit's processing element first separates packed FP data into sign, exponent, and mantissa fields using small crossbars; the Primitive Generator then forms the AND of every activation-weight mantissa bit pair, and the Flexible Bit Reduction Tree (FBRT) routes, shifts, concatenates, and adds those primitives in parallel to produce mantissa products of any width. Exponent addition runs on a segmentable carry-chain adder whose carry chains break at operand boundaries, so the same adder serves many narrow exponents or one wide exponent. The paper claims this composition keeps all compute bits active for non-power-of-two formats, eliminating both the padding waste of upcasting and the serial slowdown of bit-serial processing, with a small crossbar-based bit-packing unit condensing non-aligned data in on-chip memory.

Load-bearing premise

The reported speedups assume the cycle-accurate simulator and the modeled baselines (Tensor Core-like, BitFusion extended to FP, Cambricon-P, BitMod) predict real latency, energy, and area for cloud-scale models; the simulator is validated against RTL only for two attention-layer cases.

Editorial extensions

If this is right

  • Non-power-of-two quantization methods such as FP6-LLM could see real inference speedups on FlexiBit instead of being upcast and padded, so algorithm and hardware would finally agree on precision.
  • A single accelerator would cover power-of-two precisions, non-power-of-two precisions, and mixed operand pairs such as FP16 activation with FP4/FP5/FP6 weight, reducing the need for separate datapaths or new chip revisions when a new format appears.
  • Because the claimed performance/area advantage over bit-serial designs is 3.9x, bit-parallel flexibility is presented as scalable to cloud-scale LLM workloads where bit-serial latency becomes prohibitive.
  • With control signals generated by a compiler at layer granularity, the architecture supports per-layer precision and format switching at negligible reconfiguration cost, matching how mixed-precision quantization is typically applied to LLMs.

Reading between the lines

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

  • The paper only evaluates static per-layer precision; a natural extension the authors do not quantify is per-tile or per-token adaptive precision, where the hardware controller path they mention would be exercised and reconfiguration overhead would matter.
  • If FP6 and FP5 become standard, the Tensor Core-like baseline would need dedicated FP6 datapaths to compete, which is exactly the redesign cost FlexiBit avoids; this suggests the architecture's relative advantage grows as the set of 'standard' precisions expands.
  • The FBRT mechanism is not specific to LLMs; the same bit-parallel shift-add tree could apply to transprecision scientific computing or to accelerators for sparse or dynamic-range workloads, though the paper does not explore those settings.
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

3 major / 6 minor

Summary. The paper proposes FlexiBit, a bit-parallel accelerator architecture that aims to support arbitrary FP and INT precisions and formats without compute-unit underutilization. The key microarchitectural contribution is the Flexible Bit Reduction Tree (FBRT), which extends MAERI's augmented reduction tree to bit-level reconfigurability, together with a flexible-bit exponent adder, a concat-shift tree, and a bit-packing/unpacking unit. The authors evaluate FlexiBit in four accelerator configurations on four LLM workloads and thirteen precision pairs, reporting latency, energy, EDP, and performance-per-area improvements over a Tensor-Core-like systolic array, an FP-extended BitFusion baseline, and bit-serial baselines (Cambricon-P and BitMod). The headline claims are 1.66x and 1.62x higher performance per area on GPT-3 in FP6 versus the Tensor-Core-like and BitFusion baselines, and 3.9x versus bit-serial architectures.

Significance. If the quantitative claims are established, FlexiBit would be a significant contribution: it is, to my knowledge, the first fully flexible bit-parallel FP/INT accelerator that supports arbitrary precision and format within a register width, directly addressing the hardware-software gap highlighted by recent FP6/FP5 quantization work. The paper ships concrete evidence for FlexiBit itself: RTL implementation, post-place-and-route synthesis at 15nm NanGate, and a cycle-accurate simulator validated against RTL on attention layers of Bert-base (96%) and Llama-2-7b (99%). The architectural idea of extending MAERI-style reduction trees to bit-level operations is well motivated and the design details of FBRT, FBEA, CST, and bit-packing are described in sufficient depth to be reproduced. The main risk is not the FlexiBit design but the evaluation of competitors: the FP-extended BitFusion baseline is constructed by the authors without a disclosed model, and the headline numbers rely on simulator extrapolation to larger models and on modeled baseline energy/latency figures.

major comments (3)
  1. [§5.1, Baselines] The FP extension of BitFusion is load-bearing for the central quantitative claim (the abstract's 1.62x/1.66x performance-per-area numbers on GPT-3 FP6) but is not specified. The text states only 'we extend Bit-fusion for floating-point' and then adopts BitFusion's reported area and power, scaled by technology. BitFusion's published datapath is integer-only; supporting FP requires exponent adders, mantissa alignment, normalization, and packing of non-power-of-two mantissa widths. Without a cycle-level or RTL-level model of this extension, there is no evidence that the baseline does not, for example, pad FP6 to FP8 or assume ideal packing. Please provide the extended baseline's datapath, cycle-count formulas, and area/power derivation, or re-run the comparison against a constructed baseline whose description is sufficient for an independent reimplementation.
  2. [§5.2, Methodology and Figure 9] The cycle-accurate simulator is validated against RTL only for FlexiBit attention layers of Bert-base (96%) and Llama-2-7b (99%), yet the headline results are reported for Llama-2-70b and GPT-3 across four accelerator scales and thirteen precision pairs. The paper does not provide an error analysis or a bounded-extrapolation argument for larger models, longer sequence lengths, or different dataflow choices. Additionally, the baseline latency and energy models (Tensor-Core-like, BitFusion-like, Cambricon-P, BitMod) are not validated against any RTL or published measurements. Please either validate the simulator on the larger workloads/accuracy-critical configurations, provide a sensitivity analysis of the headline ratios to simulator error, or restrict the strongest claims to the configurations that have been validated.
  3. [§5.3.3, Tables 4 and 5, Figure 13] The EDP comparison against bit-serial baselines is sensitive to modeled latency assumptions that are not fully disclosed. FlexiBit has 3.6-4.0x larger area and higher power than Cambricon-P and BitMod (Table 5), so the EDP advantage in Table 4 and Figure 13 depends entirely on the modeled latency, e.g., the reported 52x higher latency for Cambricon-P on Llama-2-70b at Cloud-B scale. The paper does not state the clock frequency, memory tiling, or NoC assumptions used to derive Cambricon-P's latency, nor does it report a sensitivity analysis to those assumptions. Please provide the full modeling equations for Cambricon-P and BitMod and quantify how the EDP and performance-per-area conclusions change under alternative assumptions (e.g., higher clock, better tiling, or with the reported 7.1x power advantage included in a metric other than EDP).
minor comments (6)
  1. [Abstract] The abstract states '3.9x higher performance per area compared to a state-of-the-art bit-serial architecture' and then immediately refers to 'a state-of-the-art bit-parallel architecture'; the latter should read 'bit-serial' or the sentence should be rephrased.
  2. [§3.4, first paragraph] The sentence 'FBRT approaches the multiplication operation as a spatial shift-add operation and aggregates primitive bits in a tree architecture inspired by inspired by Augmented Reduction Tree (ART) of MAERI' contains a duplicated 'inspired by'.
  3. [§3.6] The description of the Exponent Normalization Unit says it 'subtracts corresponding exponents' but the surrounding text does not clarify whether this handles both the FP addition alignment case and the exponent-bias adjustment for multiplication; please clarify.
  4. [Figure 10 caption and panels] The x-axis labels in Figure 10 are dense and not fully readable at print resolution; consider separating the precision-pair labels or listing them in a table.
  5. [§5.3.1] The latency comparison text reports averages (59% and 31% less latency) but does not state whether these are geometric or arithmetic means across models and precision pairs; please specify to aid reproducibility.
  6. [Table 6] The table's row 'Power-of-two Limited flexible Bit-parallel [47]' is missing a verb or article; read as 'Power-of-two Precision/Format Preset flexible Bit-parallel [47]'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: FlexiBit's central claims rest on RTL-validated simulation, synthesized area, and independent energy estimation; no derivation step reduces to its own inputs.

full rationale

The paper's efficiency claims are produced by a cycle-accurate simulator validated against its own RTL (Section 5.2: 'The simulator is validated against RTL simulation with 96% accuracy on Bert-base and 99% accuracy on Llama-2-7b'), with PE area from Post-PnR synthesis using NanGate 15nm and energy from Accelergy. These are independent evidence rather than re-statements of the architecture's design goals. The FBRT is inspired by MAERI, a prior work of the senior author, but the paper implements FBRT in SystemVerilog and evaluates it; no conclusion is imported from MAERI by citation, and no uniqueness claim is used to forbid alternatives. The BitFusion-FP baseline is author-constructed ('we extend Bit-fusion for floating-point'), which is a modeling assumption that may affect comparison fairness, but it is not a fitted parameter renamed as a prediction, and the reported 1.62x/1.66x improvements are not algebraically forced by the baseline model's definition. The paper's own validation limitation (only FlexiBit attention layers for RTL comparison, and no detailed derivation of the FP-extended BitFusion model) is a soundness and reproducibility risk, not a circular step. No equation in the paper defines the claimed advantage in terms of its own inputs, so there is no circularity by construction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central claims rest on hand-tuned datapath parameters (reg_width=24, RM/RE=12, 144-bit datapaths) and on modeling assumptions about simulator generality, baseline fidelity, and compiler control overhead. The implicit-1 handling is standard math, not a fitted input.

free parameters (4)
  • reg_width = 24 bits
    Chosen to maximize throughput per area (Figure 14a); directly affects datapath width, crossbar sizes, and area overhead.
  • Mantissa register width RM = 12 bits
    Default design parameter from Table 1 after cost-benefit analysis; bounds the mantissa precision the PE can process in one cycle.
  • Exponent register width RE = 12 bits
    Default design parameter from Table 1; bounds exponent width and affects the flexible exponent adder size.
  • Datapath widths Lprim, LAdd, LAcc, LCST = 144 bits each
    Table 1 defaults derived from reg_width and tuned for area and performance; these widths set the maximum number of parallel primitives.
assumptions (5)
  • domain assumption FBRT switch modes and the Code 3 control algorithm implement arbitrary bit-parallel mantissa multiplication without underutilization for all supported precisions.
    Assumed from RTL validation on two attention layers; not proven across all 13 precision pairs, and [8,8]/[4,4] cases underperform TensorCore, indicating utilization is not perfectly uniform.
  • domain assumption The cycle-accurate simulator validated on Bert-base and Llama-2-7b attention layers generalizes to Llama-2-70b, GPT-3, all four scales, and all precision pairs.
    Section 5.2 reports 96%/99% validation on two models; no validation is given for the other workloads or for baseline models.
  • domain assumption The TensorCore-like and BitFusion baselines are faithfully modeled, including the authors' FP extension of BitFusion.
    Section 5.1 states BitFusion is extended by the authors for FP support; no independent validation of the baseline models is provided.
  • domain assumption Compiler-generated control signals have no runtime overhead and reconfiguration cost is under 100 cycles.
    Section 4 states control signals are stored in CSRs with negligible area; no detailed timing analysis of the reconfiguration path is given.
  • standard math The implicit-1 mantissa reconstruction using P_FBRT + (W << BWM(A)) + ({1, A} << BWM(W)) is exact for all precisions.
    This is the binomial expansion (1+mA)(1+mB)=1+mA+mB+mA*mB; it is standard arithmetic, not an ad hoc assumption.
invented entities (1)
  • None
    purpose: No new physical or conceptual entities are postulated.
    FBRT, FBEA, CST, and BPU are microarchitectural modules that are part of the design and are evaluated in RTL; they are not unexplained postulated entities like a new force or particle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FlexiBit: Fully Flexible Precision Bit-parallel Accelerator Architecture for Arbitrary Mixed Precision AI." pith.science (2026). https://pith.science/paper/RCOTM2AM

@misc{pith2026241118065,
  author       = {Pith},
  title        = {Pith review of: FlexiBit: Fully Flexible Precision Bit-parallel Accelerator Architecture for Arbitrary Mixed Precision AI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RCOTM2AM}},
  note         = {Machine review of arXiv:2411.18065}
}
read the original abstract

Recent research has shown that large language models (LLMs) can utilize low-precision floating point (FP) quantization to deliver high efficiency while maintaining original model accuracy. In particular, recent works have shown the effectiveness of non-power-of-two precisions, such as FP6 and FP5, and diverse sensitivity to low-precision arithmetic of LLM layers, which motivates mixed precision arithmetic including non-power-of-two precisions in LLMs. Although low-precision algorithmically leads to low computational overheads, such benefits cannot be fully exploited due to hardware constraints that support a limited set of power-of-two precisions (e.g., FP8, 16, 32, and 64 in NVIDIA H100 Tensor Core). In addition, the hardware compute units are designed to support standard formats (e.g., E4M3 and E5M2 for FP8). Such practices require re-designing the hardware whenever new precision and format emerge, which leads to high hardware replacement costs to exploit the benefits of new precisions and formats. Therefore, in this paper, we propose a new accelerator architecture, FlexiBit, which efficiently supports FP and INT arithmetic in arbitrary precisions and formats. Unlike previous bit-serial designs, which also provide flexibility but at the cost of performance due to its bit-wise temporal processing nature, FlexiBit's architecture enables bit-parallel processing of any precision and format without compute unit underutilization. FlexiBit's new capability to exploit non-power of two precision and format led to 1.66x and 1.62x higher performance per area on GPT-3 in FP6 targeting a cloud-scale accelerator, compared to a Tensor Core-like architecture and a state-of-the-art bit-parallel flexible precision accelerator, BitFusion, respectively. Also, the bit-parallel nature of FlexiBit's architecture led to 3.9x higher performance/area compared to a state-of-the-art bit-serial architecture.

Figures

Figures reproduced from arXiv: 2411.18065 by the authors.

Figure 1
Figure 1. Various FP quantization formats for large language models (R1: [34] R2: [50], R3: [31]) and challenges in current GPUs and accelerators for supporting such diverse FP formats. The padding example in (c) is for a positive number. common for fixed point (INT) data types. To eliminate this un￾derutilization, some works approach the problem with bit-serial architectures[15],[17], [28], [19], [6]. While bit-serial proces… view at source ↗
Figure 2
Figure 2. FlexiBit processing element (PE) architecture as LLMs. Such limitations of existing hardware architectures for large language models today with various FP precision and formats motivate us to design a new hardware architecture satisfying the following requirements: • Efficiently supporting arbitrary FP precision/formats; both in compute units and memory • A bit-parallel architecture to provide desired performance fo… view at source ↗
Figure 3
Figure 3. A walk-through operation example of FP6 activation and FP5 weight from Bit Packing unit to FBRT. For Separator in (b), we only show the weights since its operation is the same for activation. 10 s i g n _ i d x ++ 11 e l i f a c t _ b i t i d < 1 + BW_E(A) : 12 a c t _ e x p o n e n t _ r e g [ e x p o n e n t _i d x ] = a c t _ r e g [ i ] 13 e x p o n e n t _i d x ++ 14 e l s e 15 a c t _ m a n t i s s a _ r e g [… view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Microarchitecture switch and possible modes overview. Left and Right nodes refer to the tree node with neighbor link on the left and right side, respectively. The brackets "{" and "}" refer to concatenation. 3.4 FBRT: A Flexible-Bit Reduction Tree FBRT is the key enabl…
Figure 5
Figure 5. Figure 5: How FBRT handles implicit 1 for mantissa multiplication. P_FBRT refers to the intermediate results generated in FBRT. Activation Weight EA1 EA0 + EW1 EW0 * Notation: SA(i) = EAi +EWi EA2 + SA(2) SA(1) SA(0) Activation Weight EB2 EB1 EB0 + EW1 EW0 EW1 EW0 SB(1) SB(0) EB…
Figure 7
Figure 7. Figure 7: (a) The Concat-Shift Tree (CST) architecture and an ex￾ample operation on 3-bit mantissas. (b) Node microarchitecture and modes in CST. in Subsection 3.3. It subtracts corresponding exponents and identi￾fies the shift amount. 3.7 Concat-Shift Tree (CST) Using the shift…
Figure 8
Figure 8. Figure 8: Overview of FlexiBit accelerator 4 FlexiBit: Fully-flexible Precision Accelerator [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Performance model validation against RTL simulation on attention layers of Bert-base and Llama-2-7b [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Latency of Bert-Base, Llama-2-7b Llama-2-70b, and GPT-3 models on different precisions. Input sequence is 2048. Results are depicted for (a) Mobile-A configuration, (b) Mobile-B configuration, (c)Cloud-A configuration, and (d)Cloud-B configuration accelerator scales. …
Figure 11
Figure 11. Figure 11: Latency comparison of FlexiBit with and without Bit￾Packing structure. The results are normalized based on TensorCore Latency at each data precision [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: Performance per area comparison on 13 Act./Wgt. precision pairs. The sequence length is 2048. (a) Mobile-A (b) Mobile-B (c) Cloud-A (d) Cloud-B Bit Precision [W, A] Bit Precision [W, A] Bit Precision [W, A] Bit Precision [W, A] Bert-base Llama-2 (7b) Llama-2 (70b) GPT…
Figure 13
Figure 13. Figure 13: Comparison of a bit-serial (Cambricon-P and BitMod) and our bit-parallel (FlexiBit) flexible precision architecture using EDP normalized against Tensor Core-like baseline. 22% 20% 24% 23% TensorCore PE (a) Impact of Register Width on PE Area Breakdown and Throughput p…
Figure 14
Figure 14. Figure 14: The impact of register width and area breakdown. (a) PE Area breakdown of FlexiBit and throughput per area on the register width from 16 to 32 on Mobile-A scale (b) accelerator area breakdown (Reg_Width = 24) hardware utilization. BitMoD [4] introduces new precision r…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. TruncQuant: Truncation-Ready Quantization for DNNs with Flexible Weight Bit Precision

    cs.LG 2025-06 conditional novelty 6.0 of 10

    TruncQuant uses a floor-based quantizer with 2^n scaling instead of rounding with 2^n-1, so truncating high-precision weights via bit-shift exactly matches direct low-precision quantization, recovering accuracy lost b...

Reference graph

Works this paper leans on

55 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [1]

    IEEE Standard for Floating-Point Arithmetic

    2019. IEEE Standard for Floating-Point Arithmetic. IEEE Std 754-2019 (Revision of IEEE 754-2008) (2019), 1–84. https://doi.org/10.1109/IEEESTD.2019.8766229

    arXiv 2019

  2. [2]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gr...

    2020

  3. [3]

    Neil Burgess, Jelena Milanovic, Nigel Stephens, Konstantinos Monachopoulos, and David Mansell. 2019. Bfloat16 processing for neural networks. In 2019 IEEE 26th Symposium on Computer Arithmetic (ARITH) . IEEE, 88–91

    work page 2019

  4. [4]

    AbouElhamayed, Xilai Dai, Yang Wang, Marta Andronic, George A

    Yuzong Chen, Ahmed F. AbouElhamayed, Xilai Dai, Yang Wang, Marta Andronic, George A. Constantinides, and Mohamed S. Abdelfattah. 2025. BitMoD: Bit-serial Mixture-of-Datatype LLM Acceleration. In 2025 IEEE International Symposium on High Performance Computer Architecture (HPCA) . 1082–1097. https://doi.org/ 10.1109/HPCA61900.2025.00084

    arXiv 2025

  5. [5]

    Yu-Hsin Chen, Joel Emer, and Vivienne Sze. 2016. Eyeriss: A spatial architecture for energy-efficient dataflow for convolutional neural networks. In International Symposium on Computer Architecture (ISCA)

    work page 2016

  6. [6]

    Xiaoshu Cheng, Yiwen Wang, Weiran Ding, Hongfei Lou, and Ping Li. 2024. Leveraging Bit-Serial Architectures for Hardware-Oriented Deep Learning Ac- celerators with Column-Buffering Dataflow. Electronics 13, 7 (2024). https: //doi.org/10.3390/electronics13071217

    work page doi:10.3390/electronics13071217 2024

  7. [7]

    Can Cui, Yunsheng Ma, Xu Cao, Wenqian Ye, Yang Zhou, Kaizhao Liang, Jintai Chen, Juanwu Lu, Zichong Yang, Kuei-Da Liao, Tianren Gao, Erlong Li, Kun Tang, Zhipeng Cao, Tong Zhou, Ao Liu, Xinrui Yan, Shuqi Mei, Jianguo Cao, Ziran Wang, and Chao Zheng. 2024. A Survey on Multimodal Large Language Models for Autonomous Driving. In Proceedings of the IEEE/CVF W...

    2024

  8. [8]

    Alberto Delmas Lascorz, Patrick Judd, Dylan Malone Stuart, Zissis Poulos, Mostafa Mahmoud, Sayeh Sharify, Milos Nikolic, Kevin Siu, and Andreas Moshovos. 2019. Bit-tactical: A software/hardware approach to exploiting value and bit sparsity in neural networks. In Proceedings of the Twenty-Fourth Inter- national Conference on Architectural Support for Progr...

    work page 2019

Show all 55 references
  1. [9]

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition . Ieee, 248–255

    2009

  2. [10]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv prepri...

    2020 arXiv

  3. [11]

    Charles Eckert, Xiaowei Wang, Jingcheng Wang, Arun Subramaniyan, Ravi Iyer, Dennis Sylvester, David Blaaauw, and Reetuparna Das. 2018. Neural cache: Bit- serial in-cache acceleration of deep neural networks. In 2018 ACM/IEEE 45Th annual international symposium on computer arch...

    2018

  4. [12]

    Elias Frantar, Saleh Ashkboos, Torsten Hoefler, and Dan Alistarh. 2023. Gptq: Accurate post-training quantization for generative pre-trained transformers. In International Conference on Learning Representations (ICLR)

    2023

  5. [13]

    Yonggan Fu, Yang Zhao, Qixuan Yu, Chaojian Li, and Yingyan Lin. 2021. 2-in-1 Accelerator: Enabling Random Precision Switch for Winning Both Adversarial Robustness and Efficiency. In MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture (Virtual Event, Gre...

    2021

  6. [14]

    Cong Guo, Chen Zhang, Jingwen Leng, Zihan Liu, Fan Yang, Yunxin Liu, Minyi Guo, and Yuhao Zhu. 2022. Ant: Exploiting adaptive numerical data type for low-bit deep neural network quantization. In 2022 55th IEEE/ACM International Symposium on Microarchitecture (MICRO) . IEEE, 1414–1433

    2022

  7. [15]

    Yifan Hao, Yongwei Zhao, Chenxiao Liu, Zidong Du, Shuyao Cheng, Xiaqing Li, Xing Hu, Qi Guo, Zhiwei Xu, and Tianshi Chen. 2022. Cambricon-P: A Bitflow Architecture for Arbitrary Precision Computing. In 2022 55th IEEE/ACM International Symposium on Microarchitecture (MICRO) . 5...

    2022

  8. [16]

    Shen-Fu Hsiao and Yu-Hong Chen. 2020. Flexible Multi-Precision Accelera- tor Design for Deep Convolutional Neural Networks Considering Both Data Computation and Communication. In 2020 International Symposium on VLSI Design, Automation and Test (VLSI-DAT) . 1–4. https://doi.org...

    2020

  9. [17]

    Cheolwon Jo and KwangYeob Lee. 2020. Bit-Serial multiplier based Neural Processing Element with Approximate adder tree. In2020 International SoC Design Conference (ISOCC). 286–287. https://doi.org/10.1109/ISOCC50952.2020.9332993

    2020

  10. [18]

    Norm Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan, Lifeng Nai, Nishant Patil, Suvinay Subramanian, Andy Swing, Brian Towles, et al . 2023. Tpu v4: An optically reconfigurable supercomputer for machine learning with hardware support for embeddings. In Proceedings ...

    2023

  11. [19]

    Patrick Judd, Jorge Albericio, Tayler Hetherington, Tor M Aamodt, and Andreas Moshovos. 2016. Stripes: Bit-serial deep neural network computing. In 2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) . IEEE, 1–12

    2016

  12. [20]

    Hongshin Jun, Jinhee Cho, Kangseol Lee, Ho-Young Son, Kwiwook Kim, Hanho Jin, and Keith Kim. 2017. Hbm (high bandwidth memory) dram technology and architecture. In 2017 IEEE International Memory Workshop (IMW) . IEEE, 1–4

    2017

  13. [21]

    Jacob Devlin Ming-Wei Chang Kenton and Lee Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of naacL-HLT, Vol. 1. Minneapolis, Minnesota, 2

    2019

  14. [22]

    Yoongu Kim, Weikun Yang, and Onur Mutlu. 2015. Ramulator: A fast and exten- sible DRAM simulator. IEEE Computer architecture letters 15, 1 (2015), 45–49

    2015

  15. [23]

    Hyoukjun Kwon, Prasanth Chatarasi, Michael Pellauer, Angshuman Parashar, Vivek Sarkar, and Tushar Krishna. 2019. Understanding reuse, performance, and hardware cost of dnn dataflow: A data-centric approach. In Proceedings of the 52nd Annual IEEE/ACM International Symposium on ...

    2019

  16. [24]

    Hyoukjun Kwon, Liangzhen Lai, Michael Pellauer, Tushar Krishna, Yu-Hsin Chen, and Vikas Chandra. 2021. Heterogeneous Dataflow Accelerators for Multi-DNN Workloads. In IEEE International Symposium on High-Performance Computer Architecture, HPCA 2021, Seoul, South Korea, Februar...

    2021

  17. [25]

    Hyoukjun Kwon, Krishnakumar Nair, Jamin Seo, Jason Yik, Debabrata Mohapatra, Dongyuan Zhan, Jinook Song, Peter Capak, Peizhao Zhang, Peter Vajda, et al

  18. [26]

    Hyoukjun Kwon, Ananda Samajdar, and Tushar Krishna. 2018. MAERI: Enabling Flexible Dataflow Mapping over DNN Accelerators via Reconfigurable Inter- connects. In International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) . 461–475

    2018

  19. [27]

    Oseledets, and Victor S

    Vadim Lebedev, Yaroslav Ganin, Maksim Rakhuba, I. Oseledets, and Victor S. Lempitsky. 2014. Speeding-up Convolutional Neural Networks Using Fine-tuned CP-Decomposition. CoRR abs/1412.6553 (2014). https://api.semanticscholar.org/ CorpusID:15002492

    2014 arXiv

  20. [28]

    Shurui Li and Puneet Gupta. 2022. Bit-serial Weight Pools: Compression and Arbitrary Precision Execution of Neural Networks on Resource Constrained Processors. In Proceedings of Machine Learning and Systems, D. Marculescu, Y. Chi, and C. Wu (Eds.), Vol. 4. 238–250. https://pro...

    2022

  21. [29]

    Malladi, Hongzhong Zheng, Bob Brennan, and Yuan Xie

    Shuangchen Li, Dimin Niu, Krishna T. Malladi, Hongzhong Zheng, Bob Brennan, and Yuan Xie. 2017. DRISA: A DRAM-based Reconfigurable In-Situ Accelerator. In 2017 50th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO). 288–301

    2017

  22. [30]

    Ji Lin, Jiaming Tang, Haotian Tang, Shang Yang, Wei-Ming Chen, Wei- Chen Wang, Guangxuan Xiao, Xingyu Dang, Chuang Gan, and Song Han. 2024. AWQ: Activation-aware Weight Quantization for On-Device LLM Compression and Acceleration. In Proceedings of Machine Learn- ing and System...

    2024

  23. [31]

    Shih-yang Liu, Zechun Liu, Xijie Huang, Pingcheng Dong, and Kwang-Ting Cheng. 2023. LLM-FP4: 4-Bit Floating-Point Quantized Transformers. In Proceed- ings of the 2023 Conference on Empirical Methods in Natural Language Processing , Houda Bouamor, Juan Pino, and Kalika Bali (Ed...

    2023 doi

  24. [32]

    Zechun Liu, Changsheng Zhao, Forrest Iandola, Chen Lai, Yuandong Tian, Igor Fedorov, Yunyang Xiong, Ernie Chang, Yangyang Shi, Raghuraman Krishnamoor- thi, et al. 2024. Mobilellm: Optimizing sub-billion parameter language models for on-device use cases. arXiv preprint arXiv:24...

    2024 arXiv

  25. [33]

    Mayler Martins, Jody Maick Matos, Renato P Ribas, André Reis, Guilherme Schlinker, Lucio Rech, and Jens Michelsen. 2015. Open cell library in 15nm freepdk technology. In Proceedings of the 2015 Symposium on International Symposium on Physical Design. 171–178

    2015

  26. [34]

    Paulius Micikevicius, Dusan Stosic, Neil Burgess, Marius Cornea, Pradeep Dubey, Richard Grisenthwaite, Sangwon Ha, Alexander Heinecke, Patrick Judd, John Kamalu, et al. 2022. Fp8 formats for deep learning. arXiv preprint arXiv:2209.05433 (2022)

    2022 arXiv

  27. [35]

    Seock-Hwan Noh, Jahyun Koo, Seunghyun Lee, Jongse Park, and Jaeha Kung

  28. [36]

    Seock-Hwan Noh, Seungpyo Lee, Banseok Shin, Sehun Park, Yongjoo Jang, and Jaeha Kung. 2023. All-rounder: A flexible DNN accelerator with diverse data format support. arXiv preprint arXiv:2310.16757 (2023)

    2023 arXiv

  29. [37]

    IEEE Trans

    FlexBlock: A flexible DNN training accelerator with multi-mode block floating point support. IEEE Trans. Comput. 72, 9 (2023), 2522–2535

    2023

  30. [38]

    NVIDIA. 2023. NVIDIA H100 Tensor Core GPU. https://www.nvidia.com/en- us/data-center/h100/

    2023

  31. [39]

    NVIDIA. 2017. NVIDIA Tensor Cores: High-Performance Matrix Operations for Deep Learning. Technical Report. NVIDIA. https://www.nvidia.com/en-us/data- center/tensorcore/

    2017

  32. [40]

    OpenAI. 2023. GPT-4 Technical Report. arXiv preprint arXiv:2303.08774 (2023)

    2023 arXiv

  33. [41]

    NVIDIA. 2024. NVIDIA Blackwell Architecture. https://www.nvidia.com/en- us/data-center/technologies/blackwell-architecture/

    2024

  34. [42]

    Ying, Anurag Mukkara, Rangharajan Venkatesan, Brucek Khailany, Stephen W

    Angshuman Parashar, Priyanka Raina, Yakun Sophia Shao, Yu-Hsin Chen, Victor A. Ying, Anurag Mukkara, Rangharajan Venkatesan, Brucek Khailany, Stephen W. Keckler, and Joel Emer. 2019. Timeloop: A Systematic Approach to DNN Accelerator Evaluation. In 2019 IEEE International Symp...

    2019

  35. [43]

    Keckler, and William J

    Mike O’Connor, Niladrish Chatterjee, Donghyuk Lee, John Wilson, Aditya Agrawal, Stephen W. Keckler, and William J. Dally. 2017. Fine-Grained DRAM: Energy-Efficient DRAM for Extreme Bandwidth Systems. In 2017 50th Annual IEEE/ACM International Symposium on Microarchitecture (MI...

    2017

  36. [44]

    Bita Darvish Rouhani, Ritchie Zhao, Ankit More, Mathew Hall, Alireza Kho- damoradi, Summer Deng, Dhruv Choudhary, Marius Cornea, Eric Dellinger, Kristof Denolf, Stosic Dusan, Venmugil Elango, Maximilian Golub, Alexander Heinecke, Phil James-Roxby, Dharmesh Jani, Gaurav Kolhe, ...

  37. [45]

    Enrico Reggiani, Alessandro Pappalardo, Max Doblas, Miquel Moreto, Mauro Olivieri, Osman Sabri Unsal, and Adrián Cristal. 2023. Mix-GEMM: An effi- cient HW-SW Architecture for Mixed-Precision Quantized Deep Neural Net- works Inference on Edge Devices. In 2023 IEEE Internationa...

    2023

  38. [46]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yas- mine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhos- ale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 (2023)

    2023 arXiv

  39. [47]

    arXiv:2310.10537 [cs.LG] https://arxiv.org/abs/2310.10537

    Microscaling Data Formats for Deep Learning. arXiv:2310.10537 [cs.LG] https://arxiv.org/abs/2310.10537

    arXiv

  40. [48]

    Hardik Sharma, Jongse Park, Naveen Suda, Liangzhen Lai, Benson Chau, Joon Kyung Kim, Vikas Chandra, and Hadi Esmaeilzadeh. 2018. Bit fusion: Bit- level dynamically composable architecture for accelerating deep neural network. In 2018 ACM/IEEE 45th Annual International Symposiu...

    2018

  41. [49]

    Yannan Nellie Wu, Joel S Emer, and Vivienne Sze. 2019. Accelergy: An architecture-level energy estimation methodology for accelerator designs. In2019 IEEE/ACM International Conference on Computer-Aided Design (ICCAD) . IEEE, 1–8

    2019

  42. [50]

    Swagath Venkataramani, Vijayalakshmi Srinivasan, Wei Wang, Sanchari Sen, Jintao Zhang, Ankur Agrawal, Monodeep Kar, Shubham Jain, Alberto Mannari, Hoang Tran, Yulong Li, Eri Ogawa, Kazuaki Ishizaki, Hiroshi Inoue, Marcel Schaal, Mauricio Serrano, Jungwook Choi, Xiao Sun, Naiga...

    2021

  43. [51]

    Hao Wen, Yuanchun Li, Guohong Liu, Shanhui Zhao, Tao Yu, Toby Jia-Jun Li, Shiqi Jiang, Yunhao Liu, Yaqin Zhang, and Yunxin Liu. 2024. Autodroid: Llm-powered task automation in android. In Proceedings of the 30th Annual International Con- ference on Mobile Computing and Network...

    2024

  44. [53]

    Haojun Xia, Zhen Zheng, Xiaoxia Wu, Shiyang Chen, Zhewei Yao, Stephen Youn, Arash Bakhtiari, Michael Wyatt, Donglin Zhuang, Zhongzhu Zhou, et al

  45. [55]

    Jones, and Peipei Zhou

    Zhuoping Yang, Jinming Zhuang, Jiaqi Yin, Cunxi Yu, Alex K. Jones, and Peipei Zhou. 2023. AIM: Accelerating Arbitrary-Precision Integer Multiplication on Heterogeneous Reconfigurable Computing Platform Versal ACAP. In 2023 IEEE/ACM International Conference on Computer Aided De...

    2023

  46. [2023]

    Proceedings of Machine Learning and Systems 5 (2023), 1–20

    Xrbench: An extended reality (xr) machine learning benchmark suite for the metaverse. Proceedings of Machine Learning and Systems 5 (2023), 1–20

    2023

  47. [2024]

    arXiv preprint arXiv:2401.14112 (2024)

    Fp6-llm: Efficiently serving large language models through fp6-centric algorithm-system co-design. arXiv preprint arXiv:2401.14112 (2024)

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.