pith. sign in

arxiv: 2501.18253 · v2 · submitted 2025-01-30 · 💻 cs.AR

Increasing the Energy-Efficiency of Wearables Using Low-Precision Posit Arithmetic with PHEE

David Mallas\'en , Pasquale Davide Schiavone , Alberto A. Del Barrio , Manuel Prieto-Matias , David Atienza This is my paper

Pith reviewed 2026-05-23 05:06 UTC · model grok-4.3

classification 💻 cs.AR
keywords posit arithmeticwearable devicesedge AIbiomedical monitoringlow-precision computingRISC-Vcough detectionECG analysis
0
0 comments X

The pith

16-bit posit arithmetic matches 32-bit floating-point accuracy in cough detection and works with 8-10 bits for R-peak detection while shrinking hardware.

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

The paper tests posit arithmetic against IEEE 754 floating point on two biomedical edge-AI tasks: supervised cough detection and unsupervised R-peak detection in ECG signals. It finds that 16-bit posits incur only minimal accuracy loss relative to 32-bit floats for cough detection, while 8-bit or 10-bit posits already reach satisfactory accuracy for R-peak detection where floats require 16 bits. To move beyond simulation, the authors present PHEE, a RISC-V platform that embeds a posit coprocessor, and report post-synthesis results in 16 nm TSMC showing the posit units are 38 percent smaller and draw up to 42.3 percent less power at the functional-unit level with no speed penalty.

Core claim

Low-precision posit formats can replace higher-precision floating-point numbers in the supervised and unsupervised learning pipelines used for cough detection and R-peak detection, and the PHEE architecture that integrates a posit coprocessor inside a RISC-V system yields 38 percent smaller area and up to 42.3 percent lower power at the functional-unit level in 16 nm technology with no performance loss.

What carries the argument

PHEE, a modular RISC-V-based platform that embeds the Coprosit posit coprocessor and is built on the X-HEEP framework to quantify hardware-level gains of low-precision posits.

If this is right

  • 16-bit posits suffice for cough detection with only minimal accuracy loss compared with 32-bit floating point.
  • R-peak detection reaches satisfactory accuracy with 8-bit or 10-bit posits, below the 16-bit threshold required by floating point.
  • The posit functional units inside PHEE occupy 38 percent less silicon area than their floating-point counterparts.
  • Power consumption of the posit units drops by as much as 42.3 percent at the functional-unit level with no performance penalty.
  • The same accuracy can be maintained at lower bit widths, opening the possibility of reduced memory traffic in edge-AI biomedical pipelines.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If memory and interconnect costs remain modest, the reported functional-unit savings could translate into measurably longer battery life or smaller batteries in continuous-monitoring wearables.
  • The same posit substitution pattern may extend to other supervised or unsupervised biomedical signal-processing tasks that currently rely on floating-point arithmetic.
  • Lower bit widths could allow on-device models to run with reduced SRAM footprint, potentially enabling more simultaneous sensors or longer recording windows without increasing device size.

Load-bearing premise

Post-synthesis area and power numbers measured for the posit functional units alone will produce net energy savings once memory, interconnect, and other system overheads are included in a real wearable device.

What would settle it

Fabricating a complete PHEE-based wearable, running continuous cough and ECG monitoring workloads, and measuring total system energy; if the posit version shows no reduction relative to an equivalent floating-point version, the claimed efficiency gain does not hold.

Figures

Figures reproduced from arXiv: 2501.18253 by Alberto A. Del Barrio, David Atienza, David Mallas\'en, Manuel Prieto-Matias, Pasquale Davide Schiavone.

Figure 1
Figure 1. Figure 1: Posit format showing the sign, regime, exponent, and fraction fields. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Accuracy and dynamic range of 16-bit arithmetic formats. The [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: F1 score for BayeSlope executed with different arithmetic formats. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Accuracy and dynamic range of FP16, posit12, and posit10. The [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Block diagram of PHEE. Coprosit extends X-HEEP’s CPU through [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Block diagram of the PRAU. The PRAU offers a comprehensive set of features, including: • Computational operations: Addition, subtraction, multi￾plication, division, and square root for posit numbers. • Conversion operations: Transformations between posit representations and integers. • Quire accumulator operations: clearing, negation, MAC, and rounding to posit format. • Register manipulation operations: R… view at source ↗
Figure 7
Figure 7. Figure 7: Block diagram of Coprosit. The execution stage contains the PRAU, [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
read the original abstract

Wearable edge AI biomedical devices are increasingly being used for continuous patient health monitoring, enabling real-time insights and extended data collection without the need for prolonged hospital stays. These devices must be energy efficient to minimize battery size, improve comfort, and reduce recharging intervals. This paper investigates the use of specialized low-precision arithmetic formats to enhance the energy efficiency of edge AI biomedical wearables. Specifically, we explore posit arithmetic, a floating-point-like representation, in two biomedical applications that leverage supervised and unsupervised learning algorithms: cough detection for chronic cough monitoring and R peak detection in ECG analysis. Our results reveal that 16-bit posits can replace 32-bit IEEE 754 floating point numbers with minimal accuracy loss in cough detection. For R peak detection, posit arithmetic achieves satisfactory accuracy with as few as 10 or 8 bits, compared to the 16-bit requirement for floating-point formats. To validate these findings beyond algorithm-level simulations, we introduce PHEE, a modular and extensible architecture that integrates the Coprosit posit coprocessor within a RISC-V-based system. Using the X-HEEP framework, PHEE serves as a proof-of-concept platform to quantify the practical energy benefits of low-precision posits in edge AI systems. Post-synthesis results targeting 16 nm TSMC technology show that the posit hardware targeting these ML-based biomedical applications can be 38% smaller and consume up to 42.3% less power at the functional unit level, with no performance compromise. These findings establish the potential of low-precision posit arithmetic to significantly improve the energy efficiency of edge AI biomedical devices.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript claims that 16-bit posit arithmetic can substitute for 32-bit IEEE 754 floating-point with minimal accuracy loss in cough detection tasks, while 8- or 10-bit posits suffice for R-peak detection in ECG where floating-point requires 16 bits. It further presents the PHEE architecture, which integrates a posit coprocessor into an X-HEEP RISC-V platform, with post-synthesis results in 16 nm TSMC showing 38% smaller area and up to 42.3% lower power at the functional-unit level with no performance penalty.

Significance. Should the system-level energy benefits be confirmed, the work provides a concrete demonstration of posit arithmetic's utility in resource-constrained biomedical wearables, supported by hardware synthesis metrics that can serve as a baseline for future comparisons. The modular design of PHEE and the direct simulation/synthesis approach strengthen the reproducibility of the reported savings.

major comments (1)
  1. [Hardware evaluation (post-synthesis results)] The headline claim that low-precision posits 'significantly improve the energy efficiency of edge AI biomedical devices' (Abstract) depends on the functional-unit savings producing net gains at the device level. The manuscript reports only area and power for the posit coprocessor itself; no power breakdown or measurements that include the RISC-V core, memory accesses, interconnect, or sensor data movement for the target workloads are provided. In biomedical wearables, memory energy often dominates arithmetic, so the 38% / 42.3% FU-level figures may not translate to overall savings.
minor comments (2)
  1. [Abstract and evaluation sections] Accuracy results are presented without visible error bars, full dataset descriptions, or explicit comparisons against additional baselines beyond the stated floating-point widths.
  2. [PHEE architecture description] The mapping from the tested posit widths (8, 10, 16 bits) to the specific supervised/unsupervised models for cough and ECG tasks could be clarified with a table or diagram.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation of the work's significance and reproducibility, and for the constructive comment on hardware evaluation. We respond point by point below.

read point-by-point responses

  1. Referee: [Hardware evaluation (post-synthesis results)] The headline claim that low-precision posits 'significantly improve the energy efficiency of edge AI biomedical devices' (Abstract) depends on the functional-unit savings producing net gains at the device level. The manuscript reports only area and power for the posit coprocessor itself; no power breakdown or measurements that include the RISC-V core, memory accesses, interconnect, or sensor data movement for the target workloads are provided. In biomedical wearables, memory energy often dominates arithmetic, so the 38% / 42.3% FU-level figures may not translate to overall savings.

    Authors: We agree that the reported 38% area and 42.3% power reductions are measured at the functional-unit level of the posit coprocessor, as explicitly stated in the abstract and results sections. The manuscript positions PHEE as a modular proof-of-concept platform using the X-HEEP framework to demonstrate the arithmetic benefits of posits in the target biomedical workloads, without claiming full-system energy measurements. We acknowledge that memory energy can dominate in wearables and that the FU-level figures alone do not guarantee net device-level savings. In the revised manuscript we will add a dedicated discussion paragraph clarifying the scope of the claims, noting that the savings provide a concrete baseline for arithmetic energy reduction, and highlighting that the modular coprocessor design enables future system-level integration studies. We do not have additional post-synthesis data for the complete RISC-V core plus memory hierarchy at this time. revision: partial

Circularity Check

0 steps flagged

No circularity; results are direct empirical measurements

full rationale

The paper reports accuracy from running supervised/unsupervised ML algorithms on cough and ECG datasets using posit vs. float formats at varying bit widths, plus post-synthesis area/power numbers for the posit coprocessor in 16 nm TSMC via the X-HEEP framework. These are measured outputs from simulation and synthesis tools, not quantities derived by construction from fitted parameters or prior self-citations. No equations, ansatzes, or uniqueness theorems appear in the provided text; the 38% area / 42.3% power figures are synthesis results, not predictions that reduce to inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The work rests on standard hardware synthesis assumptions and the domain premise that reduced-precision arithmetic preserves task accuracy; bit widths are explored parameters rather than fitted constants; no new entities postulated.

free parameters (2)
  • posit bit widths tested (8, 10, 16)
    Specific widths chosen for comparison against float baselines in the two applications.
  • accuracy tolerance thresholds
    Implicit thresholds for what counts as minimal or satisfactory loss.
axioms (2)
  • domain assumption Low-precision posit maintains task-level accuracy for supervised and unsupervised biomedical ML models.
    Invoked when claiming 8-16 bit sufficiency replaces higher-precision float.
  • domain assumption Functional-unit synthesis metrics predict deployed energy benefits.
    Used to extrapolate from post-synthesis results to wearable system efficiency.

pith-pipeline@v0.9.0 · 5843 in / 1324 out tokens · 43171 ms · 2026-05-23T05:06:27.533660+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    A Review of Algorithm & Hardware Design for AI-Based Biomedical Applications,

    Y . Wei, J. Zhou, Y . Wang, Y . Liu, Q. Liu, J. Luo, C. Wang, F. Ren, and L. Huang, “A Review of Algorithm & Hardware Design for AI-Based Biomedical Applications,” IEEE Transactions on Biomedical Circuits and Systems , vol. 14, no. 2, pp. 145–163, Apr. 2020

    work page 2020

  2. [2]

    An Ultra- Low Power Reconfigurable Biomedical AI Processor With Adaptive Learning for Versatile Wearable Intelligent Health Monitoring,

    J. Liu, J. Fan, Z. Zhong, H. Qiu, J. Xiao, Y . Zhou, Z. Zhu, G. Dai, N. Wang, Q. Liu, Y . Xie, H. Liu, L. Chang, and J. Zhou, “An Ultra- Low Power Reconfigurable Biomedical AI Processor With Adaptive Learning for Versatile Wearable Intelligent Health Monitoring,” IEEE Transactions on Biomedical Circuits and Systems , vol. 17, no. 5, pp. 952–967, Oct. 2023

    work page 2023

  3. [3]

    Energy Harvesting For Wearable Devices: A Review,

    Y .-W. Chong, W. Ismail, K. Ko, and C.-Y . Lee, “Energy Harvesting For Wearable Devices: A Review,” IEEE Sensors Journal , vol. 19, no. 20, pp. 9047–9062, Oct. 2019

    work page 2019

  4. [4]

    INCLASS: Incremental Classification Strategy for Self- Aware Epileptic Seizure Detection,

    L. Ferretti, G. Ansaloni, R. Marquis, T. Teijeiro, P. Ryvlin, D. Atienza, and L. Pozzi, “INCLASS: Incremental Classification Strategy for Self- Aware Epileptic Seizure Detection,” in 2022 Design, Automation & Test in Europe Conference & Exhibition (DATE) . Antwerp, Belgium: IEEE, Mar. 2022, pp. 1449–1454

    work page 2022

  5. [5]

    ACE: Automated optimization towards iterative Classification in Edge health monitors,

    Y . Wang, L. Orlandic, S. Machetti, G. Ansaloni, and D. Atienza, “ACE: Automated optimization towards iterative Classification in Edge health monitors,” IEEE Transactions on Biomedical Circuits and Systems , pp. 1–11, 2024

    work page 2024

  6. [6]

    CAFS: Cost-Aware Features Selection Method for Multimodal Stress Monitoring on Wearable Devices,

    N. Momeni, A. A. Vald ´es, J. Rodrigues, C. Sandi, and D. Atienza, “CAFS: Cost-Aware Features Selection Method for Multimodal Stress Monitoring on Wearable Devices,” IEEE Transactions on Biomedical Engineering, vol. 69, no. 3, pp. 1072–1084, Mar. 2022

    work page 2022

  7. [7]

    Adaptive R- Peak Detection on Wearable ECG Sensors for High-Intensity Exercise,

    E. De Giovanni, T. Teijeiro, G. P. Millet, and D. Atienza, “Adaptive R- Peak Detection on Wearable ECG Sensors for High-Intensity Exercise,” IEEE Transactions on Biomedical Engineering , vol. 70, no. 3, pp. 941– 953, Mar. 2023

    work page 2023

  8. [8]

    ECG Arrhythmia Classification on an Ultra- Low-Power Microcontroller,

    R. Dekimpe and D. Bol, “ECG Arrhythmia Classification on an Ultra- Low-Power Microcontroller,”IEEE Transactions on Biomedical Circuits and Systems , vol. 16, no. 3, pp. 456–466, Jun. 2022. 9

    work page 2022

  9. [9]

    1.1 Computing’s energy problem (and what we can do about it),

    M. Horowitz, “1.1 Computing’s energy problem (and what we can do about it),” in 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC) , Feb. 2014, pp. 10–14

    work page 2014

  10. [10]

    The design of fast and energy-efficient linear solvers: On the potential of half-precision arithmetic and iterative refinement techniques,

    A. Haidar, A. Abdelfattah, M. Zounon, P. Wu, S. Pranesh, S. Tomov, and J. Dongarra, “The design of fast and energy-efficient linear solvers: On the potential of half-precision arithmetic and iterative refinement techniques,” in Computational Science – ICCS 2018 , Y . Shi, H. Fu, Y . Tian, V . V . Krzhizhanovskaya, M. H. Lees, J. Dongarra, and P. M. A. Slo...

    work page 2018

  11. [11]

    Standard for Posit Arithmetic (2022),

    Posit Working Group, “Standard for Posit Arithmetic (2022),” https://posithub.org/docs/posit standard-2.pdf, Feb. 2022

    work page 2022

  12. [12]

    Evaluation of the Use of Low Precision Floating-Point Arithmetic for Applications in Radio Astronomy,

    T. K. Gunaratne, “Evaluation of the Use of Low Precision Floating-Point Arithmetic for Applications in Radio Astronomy,” in Next Generation Arithmetic, J. Gustafson, S. H. Leong, and M. Michalewicz, Eds. Cham: Springer Nature Switzerland, 2023, vol. 13851, pp. 155–170

    work page 2023

  13. [13]

    Posit NPB: Assessing the Precision Improvement in HPC Scientific Applications,

    S. W. D. Chien, I. B. Peng, and S. Markidis, “Posit NPB: Assessing the Precision Improvement in HPC Scientific Applications,” in Parallel Processing and Applied Mathematics , R. Wyrzykowski, E. Deelman, J. Dongarra, and K. Karczewski, Eds. Cham: Springer International Publishing, 2020, vol. 12043, pp. 301–310

    work page 2020

  14. [14]

    Big-PERCIV AL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing,

    D. Mallas ´en, A. A. Del Barrio, and M. Prieto-Matias, “Big-PERCIV AL: Exploring the Native Use of 64-Bit Posit Arithmetic in Scientific Computing,” IEEE Transactions on Computers, vol. 73, no. 6, pp. 1472– 1485, Jun. 2024

    work page 2024

  15. [15]

    Positnn: Training Deep Neural Networks with Mixed Low-Precision Posit,

    G. Raposo, P. Tom ´as, and N. Roma, “Positnn: Training Deep Neural Networks with Mixed Low-Precision Posit,” in ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP) , Jun. 2021, pp. 7908–7912

    work page 2021

  16. [16]

    Deep PeNSieve: A deep learning framework based on the posit number system,

    R. Murillo, A. A. Del Barrio, and G. Botella, “Deep PeNSieve: A deep learning framework based on the posit number system,” Digital Signal Processing, vol. 102, p. 102762, Jul. 2020

    work page 2020

  17. [17]

    Universal Numbers Library: Design and implementation of a high-performance reproducible number systems library,

    E. T. L. Omtzigt, P. Gottschling, M. Seligman, and W. Zorn, “Universal Numbers Library: Design and implementation of a high-performance reproducible number systems library,” arXiv:2012.11011, 2020

    work page arXiv 2012

  18. [18]

    X-HEEP: An Open-Source, Configurable and Extendible RISC-V Microcontroller for the Exploration of Ultra-Low-Power Edge Accelerators,

    S. Machetti, P. D. Schiavone, T. C. M ¨uller, M. Pe ´on-Quir´os, and D. Atienza, “X-HEEP: An Open-Source, Configurable and Extendible RISC-V Microcontroller for the Exploration of Ultra-Low-Power Edge Accelerators,” Jan. 2024

    work page 2024

  19. [19]

    PERCIV AL: Open-Source Posit RISC-V Core With Quire Capability,

    D. Mallas ´en, R. Murillo, A. A. D. Barrio, G. Botella, L. Pi ˜nuel, and M. Prieto-Matias, “PERCIV AL: Open-Source Posit RISC-V Core With Quire Capability,”IEEE Transactions on Emerging Topics in Computing, vol. 10, no. 3, pp. 1241–1252, Jul. 2022

    work page 2022

  20. [20]

    Beating floating point at its own game: Posit arithmetic,

    J. L. Gustafson and I. T. Yonemoto, “Beating floating point at its own game: Posit arithmetic,”Supercomputing Frontiers and Innovations, vol. 4, no. 2, pp. 71–86, Apr. 2017

    work page 2017

  21. [21]

    Comparing Different Decodings for Posit Arithmetic,

    R. Murillo, D. Mallas ´en, A. A. Del Barrio, and G. Botella, “Comparing Different Decodings for Posit Arithmetic,” in Next Generation Arith- metic, J. Gustafson and V . Dimitrov, Eds. Cham: Springer International Publishing, 2022, vol. 13253, pp. 84–99

    work page 2022

  22. [22]

    Evaluating the Hardware Cost of the Posit Number System,

    Y . Uguen, L. Forget, and F. de Dinechin, “Evaluating the Hardware Cost of the Posit Number System,” in 2019 29th International Conference on Field Programmable Logic and Applications (FPL) . Barcelona, Spain: IEEE, Sep. 2019, pp. 106–113

    work page 2019

  23. [23]

    Fusesoc,

    Olof Kindgren, “Fusesoc,” https://github.com/olofk/fusesoc, 2025

    work page 2025

  24. [24]

    Core-V eXtension interface (CV-X-IF) v1.0.0 documentation,

    OpenHW Group, “Core-V eXtension interface (CV-X-IF) v1.0.0 documentation,” https://docs.openhwgroup.org/projects/ openhw-group-core-v-xif/en/v1.0.0/, 2024

    work page 2024

  25. [25]

    Near-threshold risc-v core with dsp extensions for scalable iot endpoint devices,

    M. Gautschi, P. D. Schiavone, A. Traber, I. Loi, A. Pullini, D. Rossi, E. Flamand, F. K. G¨urkaynak, and L. Benini, “Near-threshold risc-v core with dsp extensions for scalable iot endpoint devices,”IEEE transactions on very large scale integration (VLSI) systems , vol. 25, no. 10, pp. 2700– 2713, 2017

    work page 2017

  26. [26]

    Approximating trigono- metric functions for posits using the CORDIC method,

    J. P. Lim, M. Shachnai, and S. Nagarakatte, “Approximating trigono- metric functions for posits using the CORDIC method,” in Proceedings of the 17th ACM International Conference on Computing Frontiers . Catania Sicily Italy: ACM, May 2020, pp. 19–28

    work page 2020

  27. [27]

    Number Formats, Error Mitigation, and Scope for 16-Bit Arithmetics in Weather and Climate Modeling Analyzed With a Shallow Water Model,

    M. Kl ¨ower, P. D. D ¨uben, and T. N. Palmer, “Number Formats, Error Mitigation, and Scope for 16-Bit Arithmetics in Weather and Climate Modeling Analyzed With a Shallow Water Model,” Journal of Advances in Modeling Earth Systems , vol. 12, no. 10, p. e2020MS002246, 2020

    work page 2020

  28. [28]

    Bioformers: Embedding Transformers for Ultra-Low Power sEMG-based Gesture Recognition,

    A. Burrello, F. B. Morghet, M. Scherer, S. Benatti, L. Benini, E. Macii, M. Poncino, and D. J. Pagliari, “Bioformers: Embedding Transformers for Ultra-Low Power sEMG-based Gesture Recognition,” in 2022 De- sign, Automation & Test in Europe Conference & Exhibition (DATE) , Mar. 2022, pp. 1443–1448

    work page 2022

  29. [29]

    PERC: Posit Enhanced Rocket Chip,

    M. V . Arunkumar, S. G. Bhairathi, and H. G. Hayatnagarkar, “PERC: Posit Enhanced Rocket Chip,” in 4th Workshop on Computer Architec- ture Research with RISC-V (CARRV’20) , 2020, p. 8

    work page 2020

  30. [30]

    PERI: A Configurable Posit Enabled RISC-V Core,

    S. Tiwari, N. Gala, C. Rebeiro, and V . Kamakoti, “PERI: A Configurable Posit Enabled RISC-V Core,” ACM Transactions on Architecture and Code Optimization , vol. 18, no. 3, pp. 1–26, Jun. 2021

    work page 2021

  31. [31]

    CLARINET: A quire-enabled RISC-V- based framework for posit arithmetic empiricism,

    N. N. Sharma, R. Jain, M. M. Pokkuluri, S. B. Patkar, R. Leupers, R. S. Nikhil, and F. Merchant, “CLARINET: A quire-enabled RISC-V- based framework for posit arithmetic empiricism,” Journal of Systems Architecture, vol. 135, p. 102801, Feb. 2023

    work page 2023

  32. [32]

    A Lightweight Posit Processing Unit for RISC-V Processors in Deep Neural Network Applications,

    M. Cococcioni, F. Rossi, E. Ruffaldi, and S. Saponara, “A Lightweight Posit Processing Unit for RISC-V Processors in Deep Neural Network Applications,” IEEE Transactions on Emerging Topics in Computing , vol. 10, no. 4, pp. 1898–1908, Oct. 2022

    work page 1908

  33. [33]

    A Multimodal Dataset for Automatic Edge-AI Cough Detection,

    L. Orlandic, J. Thevenot, T. Teijeiro, and D. Atienza, “A Multimodal Dataset for Automatic Edge-AI Cough Detection,” in 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) . Sydney, Australia: IEEE, Jul. 2023, pp. 1–7

    work page 2023

  34. [34]

    BiomedBench: A bench- mark suite of TinyML biomedical applications for low-power wear- ables,

    D. Samakovlis, S. Albini, R. R. ´Alvarez, D.-A. Constantinescu, P. D. Schiavone, M. Pe ´on-Quir´os, and D. Atienza, “BiomedBench: A bench- mark suite of TinyML biomedical applications for low-power wear- ables,” IEEE Design & Test , pp. 1–1, 2024

    work page 2024

  35. [35]

    ECG in high intensity exercise dataset,

    E. De Giovanni, T. Teijeiro, D. Meier, G. Millet, and D. Atienza, “ECG in high intensity exercise dataset,” Nov. 2021

    work page 2021

  36. [36]

    FPnew: An Open-Source Multiformat Floating-Point Unit Architecture for Energy-Proportional Transprecision Computing,

    S. Mach, F. Schuiki, F. Zaruba, and L. Benini, “FPnew: An Open-Source Multiformat Floating-Point Unit Architecture for Energy-Proportional Transprecision Computing,” IEEE Transactions on V ery Large Scale Integration (VLSI) Systems , vol. 29, no. 4, pp. 774–787, Apr. 2021

    work page 2021