Pith

open record

sign in

arxiv: 2605.29533 · v1 · pith:BI76DLZV · submitted 2026-05-28 · cs.ET

Uncertainty-triggered wake-up enables energy-efficient, error-resilient edge AI with memristor front ends

Th\'eo Ballet , Aymen Romdhane , Bruno Lovison-Franco , Th\'eo Dupuis , Adrien Renaudineau , Felipe Paiva Alencar , Mohammed Akib Iftakher , Cl\'ement Turck
show 6 more authors
Kamel-Eddine Harabi Elisa Vianello Jean-Michel Portal Pascal Benoit David Novo Damien Querlioz
This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-28 23:53 UTCgrok-4.3pith:BI76DLZVrecord.jsonopen to challenge →

classification cs.ET
keywords memristorBayesian machineedge AIuncertainty estimationwake-upenergy efficiencyerror resilienceheterogeneous inference
0
0 comments X

The pith

Memristor Bayesian machine as probabilistic screener wakes CPU on uncertain outputs to preserve accuracy despite device degradation.

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

The paper establishes that a low-power memristor front end can serve as an always-on probabilistic screener in a heterogeneous inference system, triggering a higher-power CPU only when its outputs are uncertain or invalid. This converts potential silent errors from memristor variability into recoverable wake-up events, so final classification accuracy stays high even under voltage scaling or reduced programming margins. The architecture was tested on heartbeat classification by coupling a fabricated memristor Bayesian machine to an FPGA wake-up platform and CPU back end. Post-layout ASIC analysis shows average energy is set mainly by wake-up frequency, which supplies practical rules for selecting front-end operating points.

Core claim

The uncertainty-triggered wake-up architecture couples a fabricated memristor Bayesian machine to a programmable CPU, where the memristor acts as a probabilistic screener that activates the CPU via dedicated hardware on abnormal, ambiguous, or invalid outputs; the resulting system achieves high final accuracy under nominal conditions and maintains this accuracy when the memristor front end is degraded by voltage scaling or reduced programming margins because unreliable outputs become recoverable wake-up events rather than silent errors.

What carries the argument

Uncertainty-triggered wake-up path that converts unreliable memristor Bayesian outputs into CPU activations

If this is right

  • Final classification accuracy remains high under nominal operation on the heartbeat benchmark.
  • Accuracy is preserved when the memristor front end experiences voltage scaling or reduced programming margins.
  • Average system energy is governed primarily by wake-up frequency rather than constant front-end power.
  • Design rules emerge for selecting front-end operating points that balance energy and accuracy.

Where Pith is reading between the lines

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

  • The same screening principle could apply to other always-on sensing tasks where rare events justify occasional high-accuracy back-end processing.
  • Tuning the uncertainty threshold offers a direct lever to trade energy for accuracy without redesigning the memristor array.
  • If wake-up frequency stays low, the approach scales to battery-constrained devices by keeping the high-power back end dormant most of the time.

Load-bearing premise

The memristor Bayesian machine's uncertainty estimates correlate strongly enough with actual classification errors that wake-up events recover accuracy without excessive false-positive activations that erase the energy benefit.

What would settle it

Direct measurement on the heartbeat dataset showing whether the Bayesian machine's uncertainty values reliably predict its own classification errors and whether lowering programming margins increases wake-up rate while keeping final accuracy unchanged.

Figures

Figures reproduced from arXiv: 2605.29533 by Adrien Renaudineau, Aymen Romdhane, Bruno Lovison-Franco, Cl\'ement Turck, Damien Querlioz, David Novo, Elisa Vianello, Felipe Paiva Alencar, Jean-Michel Portal, Kamel-Eddine Harabi, Mohammed Akib Iftakher, Pascal Benoit, Th\'eo Ballet, Th\'eo Dupuis.

Figure 1
Figure 1. Figure 1: Heterogeneous uncertainty-triggered wake-up architecture. a Concept of the system. A memristor-based Bayesian front end performs always-on, ultra-low-power screening. Inputs confidently classified as normal are finalized locally, whereas abnormal predictions and ambiguous or invalid outputs trigger the programmable back end. b Wake-up timeline. The front end evaluates successive inputs while the programmab… view at source ↗
Figure 2
Figure 2. Figure 2: Experimental platform and characterization of the memristor front end. a Photograph of the split prototype combining the fabricated logarithmic Bayesian machine on a probe station and the remainder of the platform implemented on FPGA. b Die micrograph and functional overview of the Bayesian machine, showing the 16 memristor arrays used to accumulate class log-scores from the ECG-derived inputs and the logi… view at source ↗
Figure 3
Figure 3. Figure 3: Uncertainty-triggered wake-up restores system-level accuracy across degraded front-end conditions. a-c Confusion matrices for three representative operating conditions. VDD denotes the front-end supply voltage, and VDDR the SET-compliance voltage used during the preceding memristor-programming step. Regime A, nominal operation (VDD = 1.2 V, VDDR = 2.4 V); regime B, weaker memristor programming (VDD = 1.2 V… view at source ↗
Figure 4
Figure 4. Figure 4: Measured front-end characteristics and simulated post-layout digital platform used for system-level energy evaluation. a View of the fabricated Bayesian machine used to obtain measured front-end energy. b Measured inference energy of the Bayesian machine versus front-end supply VDD. c Post-place-and-route view of the 22-nm digital ASIC implementing the front-end support logic and programmable back end. d S… view at source ↗
Figure 5
Figure 5. Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
read the original abstract

Memristor computing offers a route to low-energy edge AI, but device variability, sensitivity to operating conditions, and system-integration challenges can hinder deployment. Here we show that these limitations can be mitigated by using memristor AI not as the final decision maker but as the ultra-low-power, always-on front end of a heterogeneous inference system. We implement this architecture by coupling a fabricated memristor Bayesian machine to a programmable CPU running a higher-power, higher-accuracy software neural network. The memristor front end acts as a probabilistic screener. When it predicts an abnormal event or produces an ambiguous or invalid output, a dedicated hardware wake-up path activates the CPU, which produces the final decision. We validate this architecture on a heartbeat-classification benchmark by interfacing the fabricated Bayesian machine with an FPGA-based wake-up platform and CPU back end. The resulting uncertainty-triggered wake-up system achieves high final classification accuracy under nominal operation and maintains this accuracy even when the memristor front end is degraded by voltage scaling or reduced programming margins, because unreliable outputs are converted into recoverable wake-up events instead of becoming silent errors. Post-layout analysis of an ASIC implementation shows that average energy is governed primarily by wake-up frequency, providing practical design rules for choosing front-end operating points. These results establish uncertainty-triggered wake-up as a strategy for energy-efficient, error-resilient edge AI.

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

3 major / 1 minor

Summary. The paper proposes an uncertainty-triggered wake-up architecture in which a fabricated memristor Bayesian machine operates as an ultra-low-power, always-on front end that detects ambiguous, invalid, or abnormal outputs and activates a higher-power CPU back-end via a dedicated wake-up path for final inference. This heterogeneous system is demonstrated on a heartbeat-classification benchmark interfaced with an FPGA platform, with the claim that final accuracy remains high under nominal conditions and is preserved under memristor degradation (voltage scaling or reduced programming margins) because silent errors are converted to recoverable wake-ups. Post-layout ASIC analysis indicates that average energy consumption is governed primarily by wake-up frequency, yielding practical design rules for front-end operating points.

Significance. If the experimental correlation between memristor uncertainty estimates and actual classification errors proves sufficiently strong, the approach offers a concrete strategy for deploying variable, condition-sensitive memristor devices in energy-constrained edge AI without requiring perfect device-level reliability, while providing quantitative guidance on energy versus accuracy trade-offs via wake-up frequency.

major comments (3)
  1. [Abstract / validation section] Abstract and validation results: the central claim that accuracy is maintained under voltage scaling and reduced margins because unreliable outputs become recoverable wake-ups is not supported by any reported quantitative metrics (accuracy-vs-degradation curves with/without wake-up, wake-up frequency vs. operating point, uncertainty calibration plots, or error-detection AUC); without these data the required correlation strength between uncertainty estimates and actual errors cannot be evaluated.
  2. [Abstract / results] Abstract and results: no baseline comparisons (e.g., direct memristor inference without wake-up, or alternative uncertainty thresholds) or error bars on the reported accuracy figures are provided, leaving the energy benefit and robustness claims unquantified relative to conventional approaches.
  3. [Abstract] Abstract: the statement that uncertainty thresholds enable the observed resilience provides no description of how those thresholds were selected or tuned, which is load-bearing for reproducing the energy-accuracy operating points claimed in the ASIC analysis.
minor comments (1)
  1. [Abstract] The abstract refers to 'abnormal event' detection but does not clarify whether this uses the same uncertainty metric as the ambiguous/invalid output case or a separate mechanism.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments, which have helped us identify areas where the presentation of our results can be strengthened. We address each major comment below and indicate the corresponding revisions to the manuscript.

read point-by-point responses

  1. Referee: [Abstract / validation section] Abstract and validation results: the central claim that accuracy is maintained under voltage scaling and reduced margins because unreliable outputs become recoverable wake-ups is not supported by any reported quantitative metrics (accuracy-vs-degradation curves with/without wake-up, wake-up frequency vs. operating point, uncertainty calibration plots, or error-detection AUC); without these data the required correlation strength between uncertainty estimates and actual errors cannot be evaluated.

    Authors: We agree that explicit quantitative metrics are needed to substantiate the central claim. The full manuscript contains supporting experimental data from the fabricated device and FPGA platform, but these were not presented as the requested curves or AUC values. In the revised manuscript we have added accuracy-versus-degradation curves (with and without wake-up), wake-up frequency versus operating point, uncertainty calibration plots, and error-detection AUC results. These additions directly quantify the correlation between uncertainty estimates and classification errors under voltage scaling and reduced programming margins. revision: yes

  2. Referee: [Abstract / results] Abstract and results: no baseline comparisons (e.g., direct memristor inference without wake-up, or alternative uncertainty thresholds) or error bars on the reported accuracy figures are provided, leaving the energy benefit and robustness claims unquantified relative to conventional approaches.

    Authors: We accept that baseline comparisons and statistical error bars would better quantify the energy and robustness benefits. The original experiments included repeated measurements, but these were not reported with error bars or explicit baselines. The revised manuscript now includes (i) direct comparison to memristor-only inference without wake-up, (ii) results for alternative uncertainty thresholds, and (iii) error bars on all accuracy figures derived from multiple device runs and FPGA trials. revision: yes

  3. Referee: [Abstract] Abstract: the statement that uncertainty thresholds enable the observed resilience provides no description of how those thresholds were selected or tuned, which is load-bearing for reproducing the energy-accuracy operating points claimed in the ASIC analysis.

    Authors: The threshold selection procedure is described in the methods section of the original manuscript, where thresholds were chosen on a held-out validation set to achieve a target wake-up rate while preserving accuracy. To address the referee's concern, we have expanded the abstract and results section with a concise description of the tuning process and the specific validation criterion used. The ASIC energy estimates remain unchanged because they are governed by the resulting wake-up frequency, which is now explicitly linked to the threshold choice. revision: partial

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper describes an experimental hardware architecture coupling a fabricated memristor Bayesian machine to a CPU back-end for uncertainty-triggered wake-up on a heartbeat-classification task. All central claims rest on measured accuracy under nominal and degraded conditions (voltage scaling, reduced margins) plus post-layout ASIC energy analysis, with no equations, parameter fits presented as predictions, or derivation chains that reduce to inputs by construction. The provided text contains no self-citation load-bearing steps, ansatz smuggling, or renaming of known results; the result is therefore self-contained against external fabrication and testing benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated.

pith-pipeline@v0.9.1-grok · 5846 in / 1103 out tokens · 25328 ms · 2026-06-28T23:53:12.607426+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

40 extracted references · 1 canonical work pages

  1. [1]

    & Wong, H.-S

    Ielmini, D. & Wong, H.-S. P. In-memory computing with resistive switching devices.Nat. Electron. 1, 333 (2018)

    2018

  2. [2]

    & Strukov, D

    Kim, H., Mahmoodi, M., Nili, H. & Strukov, D. B. 4k-memristor analog-grade passive crossbar circuit.Nat. communications12, 5198 (2021)

    2021

  3. [3]

    B., Zhang, W., Wu, D., Deiss, S., Raina, P., Qian, H., Gao, B.et al.A compute-in-memory chip based on resistive random-access memory

    Wan, W., Kubendran, R., Schaefer, C., Eryilmaz, S. B., Zhang, W., Wu, D., Deiss, S., Raina, P., Qian, H., Gao, B.et al.A compute-in-memory chip based on resistive random-access memory. Nature608, 504–512 (2022). 23/27

    2022

  4. [4]

    Electron.4, 921–930 (2021)

    Hung, J.-M., Xue, C.-X., Kao, H.-Y., Huang, Y.-H., Chang, F.-C., Huang, S.-P., Liu, T.-W., Jhang, C.-J., Su, C.-I., Khwa, W.-S.et al.A four-megabit compute-in-memory macro with eight-bit precision based on cmos and resistive random-access memory for ai edge devices.Nat. Electron.4, 921–930 (2021)

    2021

  5. [5]

    Huang, Y., Ando, T., Sebastian, A., Chang, M.-F., Yang, J. J. & Xia, Q. Memristor-based hardware accelerators for artificial intelligence.Nat. Rev. Electr. Eng.1–14 (2024)

    2024

  6. [6]

    & Strachan, J

    Li, C., Ignowski, J., Sheng, X., Wessel, R., Jaffe, B., Ingemi, J., Graves, C. & Strachan, J. P. Cmos-integrated nanoscale memristive crossbars for cnn and optimization acceleration. In2020 IEEE International Memory Workshop (IMW), 1–4 (IEEE, 2020)

    2020

  7. [7]

    D., Ramu, K., Fream, M., Borders, W

    Yousuf, O., Hoskins, B. D., Ramu, K., Fream, M., Borders, W. A., Madhavan, A., Daniels, M. W., Dienstfrey, A., McClelland, J. J., Lueker-Boden, M.et al.Layer ensemble averaging for fault tolerance in memristive neural networks.Nat. Commun.16, 1250 (2025)

    2025

  8. [8]

    Commun.15, 741 (2024)

    Jebali, F., Majumdar, A., Turck, C., Harabi, K.-E., Faye, M.-C., Muhr, E., Walder, J.-P., Bilousov, O., Michaud, A., Vianello, E.et al.Powering ai at the edge: A robust, memristor-based binarized neural network with near-memory computing and miniaturized solar cell.Nat. Commun.15, 741 (2024)

    2024

  9. [9]

    et al.Uncertainty quantification via a memristor bayesian deep neural network for risk-sensitive reinforcement learning.Nat

    Lin, Y., Zhang, Q., Gao, B., Tang, J., Yao, P., Li, C., Huang, S., Liu, Z., Zhou, Y., Liu, Y. et al.Uncertainty quantification via a memristor bayesian deep neural network for risk-sensitive reinforcement learning.Nat. Mach. Intell.5, 714–723 (2023)

    2023

  10. [10]

    Commun.14, 7530 (2023)

    Bonnet, D., Hirtzlin, T., Majumdar, A., Dalgaty, T., Esmanhotto, E., Meli, V., Castellani, N., Martin, S., Nodin, J.-F., Bourgeois, G.et al.Bringing uncertainty quantification to the extreme-edge with memristor-based bayesian neural networks.Nat. Commun.14, 7530 (2023)

    2023

  11. [11]

    Lin, Y., Gao, B., Tang, J., Zhang, Q., Qian, H. & Wu, H. Deep bayesian active learning using in-memory computing hardware.Nat. Comput. Sci.5, 27–36 (2025)

    2025

  12. [12]

    J., Beausoleil, R., Lu, W

    Cai, F., Kumar, S., Van Vaerenbergh, T., Sheng, X., Liu, R., Li, C., Liu, Z., Foltin, M., Yu, S., Xia, Q., Yang, J. J., Beausoleil, R., Lu, W. D. & Strachan, J. P. Power-efficient combinatorial optimization using intrinsic noise in memristor hopfield neural networks.Nat. Electron.3, 409–418 (2020)

    2020

  13. [13]

    Jiang, M., Shan, K., Sheng, X., Graves, C., Strachan, J. P. & Li, C. An efficient synchronous- updating memristor-based ising solver for combinatorial optimization. In2022 International Electron Devices Meeting (IEDM), 22.2.1–22.2.4 (IEEE, 2022)

    2022

  14. [14]

    Jiang, M., Shan, K., He, C. & Li, C. Efficient combinatorial optimization by quantum-inspired parallel annealing in analogue memristor crossbar.Nat. communications14, 5927 (2023)

    2023

  15. [15]

    & Kim, H

    Kim, K., Youn, S., Park, J. & Kim, H. Ising solver using weight profile of memristor crossbar array for combinatorial optimization. In2024 IEEE International Electron Devices Meeting (IEDM), 1–4 (IEEE, 2024). 24/27

    2024

  16. [16]

    One-step combinatorial optimization solver with fully integrated analog memristors and annealing module

    Shan, K., Jiang, M., Du, Z., Xiao, Y., Tong, Y., Li, Z., Yang, S.-H., He, C., Mao, R., Lin, P.et al. One-step combinatorial optimization solver with fully integrated analog memristors and annealing module. In2024 IEEE International Electron Devices Meeting (IEDM), 1–4 (IEEE, 2024)

    2024

  17. [17]

    Iftakher, M. A., Levices, H., Harabi, K.-E., Renaudineau, A., Faye, M.-C., Bouchard, C., Disdier, F., Viala, B., Vianello, E., Talatchian, P.et al.Intrinsic annealing in a hybrid memristor-magnetic tunnel junction ising machine.Nat. Commun.(2026)

    2026

  18. [18]

    & Querlioz, D

    Harabi, K.-E., Hirtzlin, T., Turck, C., Vianello, E., Laurent, R., Droulez, J., Bessière, P., Portal, J.-M., Bocquet, M. & Querlioz, D. A memristor-based bayesian machine.Nat. Electron.6, 52–63 (2023)

    2023

  19. [19]

    Eng.4, 35 (2025)

    Turck, C., Harabi, K.-E., Pontlevy, A., Ballet, T., Hirtzlin, T., Vianello, E., Laurent, R., Droulez, J., Bessière, P., Bocquet, M.et al.The logarithmic memristor-based bayesian machine.Commun. Eng.4, 35 (2025)

    2025

  20. [20]

    Ambrogio, S., Narayanan, P., Okazaki, A., Fasoli, A., Mackin, C., Hosokawa, K., Nomura, A., Yasuda, T., Chen, A., Friz, A.et al.An analog-ai chip for energy-efficient speech recognition and transcription.Nature620, 768–775 (2023)

    2023

  21. [21]

    M.et al.A 64-core mixed-signal in-memory compute chip based on phase-change memory for deep neural network inference.Nat

    Le Gallo, M., Khaddam-Aljameh, R., Stanisavljevic, M., Vasilopoulos, A., Kersting, B., Dazzi, M., Karunaratne, G., Brändli, M., Singh, A., Mueller, S. M.et al.A 64-core mixed-signal in-memory compute chip based on phase-change memory for deep neural network inference.Nat. Electron.6, 680–693 (2023)

    2023

  22. [22]

    A., Redwing, J

    Sebastian, A., Pendurthi, R., Kozhakhmetov, A., Trainor, N., Robinson, J. A., Redwing, J. M. & Das, S. Two-dimensional materials-based probabilistic synapses and reconfigurable neurons for measuring inference uncertainty using bayesian neural networks.Nat. communications13, 6139 (2022)

    2022

  23. [23]

    & Lee, J.-S

    Kim, M.-K., Kim, I.-J. & Lee, J.-S. Cmos-compatible compute-in-memory accelerators based on integrated ferroelectric synaptic arrays for convolution neural networks.Sci. Adv.8, eabm8537 (2022)

    2022

  24. [24]

    K., Kwon, S.-W., Ju, Y., Kim, M., Yi, W., Han, S

    Jung, S., Lee, H., Myung, S., Kim, H., Yoon, S. K., Kwon, S.-W., Ju, Y., Kim, M., Yi, W., Han, S. et al.A crossbar array of magnetoresistive memory devices for in-memory computing.Nature601, 211–216 (2022)

    2022

  25. [25]

    & Benini, L

    Di Mauro, A., Scherer, M., Rossi, D. & Benini, L. Kraken: A direct event/frame-based multi-sensor fusion soc for ultra-efficient visual processing in nano-uavs.arXiv preprint arXiv:2209.01065 (2022)

    work page arXiv 2022

  26. [26]

    A., Jung, V

    Busia, P., Scrugli, M. A., Jung, V. J.-B., Benini, L. & Meloni, P. A tiny transformer for low-power arrhythmia classification on microcontrollers.IEEE Transactions on Biomed. Circuits Syst.19, 142–152 (2024)

    2024

  27. [27]

    D., Conti, F., Rossi, D., Gautschi, M., Pullini, A., Flamand, E

    Schiavone, P. D., Conti, F., Rossi, D., Gautschi, M., Pullini, A., Flamand, E. & Benini, L. Slow and steady wins the race? a comparison of ultra-low-power risc-v cores for internet-of-things 25/27 applications. In2017 27th International Symposium on Power and Timing Modeling, Optimization and Simulation (PATMOS), 1–8 (IEEE, 2017)

    2017

  28. [28]

    S., Das, S., Pant, S., Rao, R., Pham, T., Ziesler, C., Blaauw, D., Austin, T., Flautner, K.et al.Razor: A low-power pipeline based on circuit-level timing speculation

    Ernst, D., Kim, N. S., Das, S., Pant, S., Rao, R., Pham, T., Ziesler, C., Blaauw, D., Austin, T., Flautner, K.et al.Razor: A low-power pipeline based on circuit-level timing speculation. InProceedings. 36th Annual IEEE/ACM International Symposium on Microarchitecture, 2003. MICRO-36., 7–18 (IEEE, 2003)

    2003

  29. [29]

    & Benini, L

    Rovere, G., Fateh, S. & Benini, L. A 2.2-muw cognitive always-on wake-up circuit for event-driven duty-cycling of iot sensor nodes.IEEE J. on Emerg. Sel. Top. Circuits Syst.8, 543–554 (2018)

    2018

  30. [30]

    Circuits Syst.9, 641–651 (2015)

    Izumi, S., Yamashita, K., Nakano, M., Yoshimoto, S., Nakagawa, T., Nakai, Y., Kawaguchi, H., Kimura, H., Marumoto, K., Fuchikami, T.et al.Normally off ecg soc with non-volatile mcu and noise tolerant heartbeat detector.IEEE Transactions on Biomed. Circuits Syst.9, 641–651 (2015)

    2015

  31. [31]

    & Thakur, C

    Gupta, S., Kumar, P., Paul, T., van Schaik, A., Ghosh, A. & Thakur, C. S. Low power, cmos-mos2 memtransistor based neuromorphic hybrid architecture for wake-up systems.Sci. reports9, 15604 (2019)

    2019

  32. [32]

    R., Chatterjee, O., Paul, T., Ghosh, A., Chakrabartty, S

    Kumar, P., Nair, A. R., Chatterjee, O., Paul, T., Ghosh, A., Chakrabartty, S. & Thakur, C. S. Neuromorphic in-memory computing framework using memtransistor cross-bar based support vector machines. In2019 IEEE 62nd International Midwest Symposium on Circuits and Systems (MWSCAS), 311–314 (IEEE, 2019)

    2019

  33. [33]

    & Vianello, E

    Querlioz, D. & Vianello, E. Bayesian electronics for trustworthy artificial intelligence.Nat. Rev. Electr. Eng.1–10 (2025)

    2025

  34. [34]

    & Noziere, J.-P

    Zhao, W., Chappert, C., Javerliac, V. & Noziere, J.-P. High speed, high stability and low power sensing amplifier for mtj/cmos hybrid logic circuits.IEEE Transactions on Magn.45, 3784–3787 (2009)

    2009

  35. [35]

    I: Regul

    Zhao, W., Moreau, M., Deng, E., Zhang, Y., Portal, J.-M., Klein, J.-O., Bocquet, M., Aziza, H., Deleruyelle, D., Muller, C.et al.Synchronous non-volatile logic gate design based on resistive switching memories.IEEE Transactions on Circuits Syst. I: Regul. Pap.61, 443–454 (2014)

    2014

  36. [36]

    L., Guilhot, Y., Miquel, J., Soriano, T., Novo, D

    Paiva-Alencar, F., Romdhane, A., Franco, B. L., Guilhot, Y., Miquel, J., Soriano, T., Novo, D. & Benoit, P. Adam: Adaptive microcontroller platform for edge ai systems. In36th International Workshop on Rapid System Prototyping (RSP’25)(2025)

    2025

  37. [37]

    & Querlioz, D

    Hirtzlin, T., Bocquet, M., Penkovsky, B., Klein, J.-O., Nowak, E., Vianello, E., Portal, J.-M. & Querlioz, D. Digital biologically plausible implementation of binarized neural networks with differential hafnium oxide resistive memory arrays.Front. neuroscience13, 1383 (2020)

    2020

  38. [38]

    & Goldberger, A

    Moody, G., Mark, R. & Goldberger, A. Physionet: a research resource for studies of complex physiologic and biomedical signals. InComputers in Cardiology 2000. Vol. 27 (Cat. 00CH37163), 179–182 (IEEE, 2000)

    2000

  39. [39]

    machine learning systems3, 800–811 (2021)

    David, R., Duke, J., Jain, A., Janapa Reddi, V., Jeffries, N., Li, J., Kreeger, N., Nappier, I., Natraj, M., Wang, T.et al.Tensorflow lite micro: Embedded machine learning for tinyml systems.Proc. machine learning systems3, 800–811 (2021). 26/27

    2021

  40. [40]

    Inci, A., Isgenc, M. M. & Marculescu, D. Deepnvm++: Cross-layer modeling and optimization framework of nonvolatile memories for deep learning.IEEE Transactions on Comput. Des. Integr. Circuits Syst.41, 3426–3437 (2021). 27/27

    2021