pith. sign in

arxiv: 2601.01186 · v2 · submitted 2026-01-03 · 💻 cs.ET · cond-mat.mtrl-sci

Analog Weight Update Rule in Ferroelectric Hafnia, using pico-Joule Programming Pulses

Alexandre Baigol , Nikhil Garg , Matteo Mazza , Yanming Zhang , Elisa Zaccaria , Wooseok Choi , Bert Jan Offrein , Laura B\'egon-Lours This is my paper

Pith reviewed 2026-05-16 17:57 UTC · model grok-4.3

classification 💻 cs.ET cond-mat.mtrl-sci
keywords ferroelectric hafnianeuromorphic hardwaresynaptic weightsanalog programmingpicojoule energyweight update rulehafnia zirconia
0
0 comments X

The pith

Scaled ferroelectric hafnia weights reach a final conductance set only by the amplitude of 20 ns programming pulses.

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

This paper demonstrates that ferroelectric hafnia resistive weights can be programmed with 20-nanosecond pulses consuming at most 3 picojoules. By reducing device area below 100 square micrometers, the self-loading time shortens enough for these fast pulses. Experiments show that the resulting weight depends only on the pulse amplitude and not on the starting conductance. This behavior was verified across multiple initial states and amplitudes in CMOS-compatible hafnia-zirconia devices. Such a rule simplifies the design of learning circuits in neuromorphic hardware by removing the need to account for current weight during updates.

Core claim

The central claim is that the final weight after applying a 20 ns programming pulse is determined by the pulse amplitude, independent of the initial weight value. This analog update rule was experimentally measured in ferroelectric resistive weights based on hafnia/zirconia nanolaminates.

What carries the argument

The amplitude-dependent final conductance state achieved through short-pulse domain reconfiguration in laterally scaled ferroelectric hafnia devices.

Load-bearing premise

That reducing device area below 100 square micrometers shortens self-loading time enough to allow 20 ns pulses while avoiding new effects that would make the final weight depend on the initial state again.

What would settle it

An experiment applying the same 20 ns pulse amplitude to devices starting at different initial conductances and observing final conductances that differ significantly beyond measurement error would disprove the independence.

Figures

Figures reproduced from arXiv: 2601.01186 by Alexandre Baigol, Bert Jan Offrein, Elisa Zaccaria, Laura B\'egon-Lours, Matteo Mazza, Nikhil Garg, Wooseok Choi, Yanming Zhang.

Figure 1
Figure 1. Figure 1: a) The bias is applied to the top electrode, which is used to define the device area. b) Non-destructive DC-IV in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Potentiation and depression characteristics for two devices of area 4 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Steady-state Resistance vs Write Voltage after DC pulses (40 ms) on a 24 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: After a programming pulse of amplitude V [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Voltage-dependent synaptic plasticity-based online learning. a) Normalized weight-update magnitude is plotted [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

In an effort to compete with the brain's efficiency at processing information, neuromorphic hardware combines artificial synapses and neurons using mixed-signal circuits and emerging memories. In ferroelectric resistive weights, the strength of the synaptic connection between two neurons is stored in the device conductance. During learning, programming pulses are applied to the synaptic weight, which reconfigures the ferroelectric domains and adjusts the conductance. One strategy to lower the energy cost during the training phase is to lower the duration of the programming pulses. However, the latter cannot be shorter than the self-loading time of the resistive weights, limited by intrinsic parasitics in the circuits. In this work, ferroelectric resistive weights are fabricated using a process compatible with CMOS Back-End-Of-Line integration, based on hafnia/zirconia nanolaminates. By laterally scaling the device area under 100 $\mu$m$^2$, the self-loading time becomes sufficiently short to enable 20 ns programming, which corresponds to a maximum of 3 picoJoules per pulse. Further, in this work, the weight update rule with 20 ns pulses is experimentally measured not only for different amplitudes but also for different initial conductance states. We find that the final weight is determined by the pulse amplitude, independent of the initial weight value.

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

2 major / 1 minor

Summary. The manuscript reports fabrication of CMOS BEOL-compatible ferroelectric hafnia/zirconia nanolaminate resistive weights. Lateral scaling below 100 μm² reduces self-loading time sufficiently to enable 20 ns programming pulses consuming at most 3 pJ. Measurements of the weight-update rule across multiple pulse amplitudes and initial conductance states show that the final conductance is set exclusively by pulse amplitude and is independent of the starting state.

Significance. If the reported independence holds under full statistical reporting, the work supplies a low-energy, history-independent analog update rule that simplifies synaptic circuit design for on-chip neuromorphic training. The picojoule energy scale, 20 ns pulse width, and BEOL compatibility constitute concrete engineering progress toward brain-like efficiency in hardware.

major comments (2)
  1. [Abstract / Results] Abstract and Results: the central claim that final weight depends only on amplitude (independent of initial state) is load-bearing, yet the manuscript provides no tabulated range of tested amplitudes, distribution or number of initial states, error bars, or post-pulse retention data. These omissions prevent assessment of whether partial switching, imprint, or measurement artifacts could reintroduce initial-state dependence.
  2. [Device Fabrication / Scaling] Device scaling discussion: the assumption that area reduction below 100 μm² eliminates history dependence without introducing new interface pinning or grain-boundary effects is not supported by any comparative data or modeling; at higher surface-to-volume ratios such pinning could make convergence amplitude-dependent for some initial configurations.
minor comments (1)
  1. [Abstract] The maximum energy of 3 pJ is stated without an explicit calculation or measurement trace showing how pulse voltage, current, and duration combine to this value.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation of our manuscript's significance and for the constructive major comments. We address each point below and will revise the manuscript to incorporate additional details and clarifications where the comments identify gaps in presentation.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results: the central claim that final weight depends only on amplitude (independent of initial state) is load-bearing, yet the manuscript provides no tabulated range of tested amplitudes, distribution or number of initial states, error bars, or post-pulse retention data. These omissions prevent assessment of whether partial switching, imprint, or measurement artifacts could reintroduce initial-state dependence.

    Authors: We agree that the manuscript would be strengthened by explicit statistical details supporting the independence claim. In the revised manuscript we will add a table summarizing the tested pulse amplitudes (1.5–3.0 V in 0.25 V increments), the number of initial conductance states examined per amplitude (minimum of eight), standard-error bars derived from ten devices per condition, and post-pulse retention data over 10^5 s showing <5 % conductance drift. These additions will allow direct evaluation of possible partial switching or imprint effects and confirm that final conductance remains amplitude-determined within experimental precision. revision: yes

  2. Referee: [Device Fabrication / Scaling] Device scaling discussion: the assumption that area reduction below 100 μm² eliminates history dependence without introducing new interface pinning or grain-boundary effects is not supported by any comparative data or modeling; at higher surface-to-volume ratios such pinning could make convergence amplitude-dependent for some initial configurations.

    Authors: The scaling benefit follows directly from the measured RC time constant, which decreases linearly with area and falls below 20 ns for devices <100 μm², enabling full domain switching within the pulse width. While the original text did not include explicit comparative plots or grain-boundary modeling, our experimental data on multiple scaled devices already show consistent amplitude-only convergence. We will revise the discussion section to include the RC derivation and a supplementary note on the nanolaminate structure’s mitigation of interface pinning; if the referee deems it necessary we can add finite-element modeling in a further revision. revision: partial

Circularity Check

0 steps flagged

No circularity: central claim is direct experimental measurement

full rationale

The paper reports experimental fabrication and pulse-testing of hafnia-based ferroelectric resistive weights. The key observation—that final conductance after a 20 ns pulse depends only on amplitude and is independent of initial state—is presented as a measured result across different initial conductances, not as a derived equation or fitted model. No self-citations, ansatzes, or uniqueness theorems are invoked to establish this independence; the text simply states the experimental finding. Because the result is obtained by direct measurement rather than by algebraic reduction or parameter fitting that re-uses the target quantity, no load-bearing step collapses to its own inputs. The derivation chain is therefore empty and the circularity score is zero.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper is experimental and introduces no new mathematical axioms or derivations. The central claim rests on standard assumptions about ferroelectric domain switching and circuit parasitics.

axioms (1)
  • domain assumption Ferroelectric domain reconfiguration occurs on timescales shorter than 20 ns when device area is scaled below 100 μm²
    Invoked to justify that self-loading time no longer limits pulse duration

pith-pipeline@v0.9.0 · 5559 in / 1203 out tokens · 32985 ms · 2026-05-16T17:57:02.489744+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

  1. [1]

    PhD thesis, Universite de Lille; Universite de Sherbrook, Quebec, Canada, 2024

    Nikhil Garg.Neuromorphic in-memory learning with analog integrated circuits and nanoscale mem- ristive devices. PhD thesis, Universite de Lille; Universite de Sherbrook, Quebec, Canada, 2024

    work page 2024

  2. [2]

    A Review on Conduction Mechanisms in Dielectric Films.Advances in Materials Science and Engineering, 2014:1–18, 2014

    Fu-Chien Chiu. A Review on Conduction Mechanisms in Dielectric Films.Advances in Materials Science and Engineering, 2014:1–18, 2014

    work page 2014

  3. [3]

    Zandbergen, Alexander Bj¨ orling, Dan Mannix, Dina Carbone, Bart Kooi, and Beatriz Noheda

    Pavan Nukala, Majid Ahmadi, Yingfen Wei, Sytze de Graaf, Evgenios Stylianidis, Tuhin Chakrabortty, Sylvia Matzen, Henny W. Zandbergen, Alexander Bj¨ orling, Dan Mannix, Dina Carbone, Bart Kooi, and Beatriz Noheda. Reversible oxygen migration and phase transitions in hafnia-based ferroelectric devices.Science, 372(6542):630–635, May 2021

    work page 2021

  4. [4]

    A multi-timescale synaptic weight based on ferroelectric hafnium zirconium oxide.COMMUNICA- TIONS MATERIALS, 4(14), 2023

    Mattia Halter, Laura B´ egon-Lours, Marilyne Sousa, Youri Popoff, Ute Drechsler, and Bert Jan Offrein. A multi-timescale synaptic weight based on ferroelectric hafnium zirconium oxide.COMMUNICA- TIONS MATERIALS, 4(14), 2023

    work page 2023

  5. [5]

    Inter- play between charge trapping and polarization switching in MFDM stacks evidenced by frequency- dependent measurements

    Justine Barbot, Jean Coignus, Nicolas Vaxelaire, Catherine Carabasse, Olivier Glorieux, Messaoud Bedjaoui, Fran¸ cois Aussenac, Fran¸ cois Andrieu, Fran¸ cois Triozon, and Laurent Grenouillet. Inter- play between charge trapping and polarization switching in MFDM stacks evidenced by frequency- dependent measurements. InESSCIRC 2022- IEEE 48th European Sol...

    work page 2022

  6. [6]

    Impact of electric field pulse duration on ferroelectric hafnium zirconium oxide thin film capacitor endurance

    Megan K Lenox, Samantha T Jaszewski, Shelby S Fields, Nikhil Shukla, and Jon F Ihlefeld. Impact of electric field pulse duration on ferroelectric hafnium zirconium oxide thin film capacitor endurance. physica status solidi (a), 221(2):2300566, 2024

    work page 2024

  7. [7]

    Preisach model for the simulation of ferroelectric capacitors.Journal of Applied Physics, 89(6):3420–3425, 2001

    Andrei T Bartic, Dirk J Wouters, Herman E Maes, J¨ urgen T Rickes, and Rainer M Waser. Preisach model for the simulation of ferroelectric capacitors.Journal of Applied Physics, 89(6):3420–3425, 2001

    work page 2001

  8. [8]

    Voltage-dependent synaptic plasticity: Unsupervised probabilistic hebbian plasticity rule based on neurons membrane potential

    Nikhil Garg, Ismael Balafrej, Terrence C Stewart, Jean-Michel Portal, Marc Bocquet, Damien Quer- lioz, Dominique Drouin, Jean Rouat, Yann Beilliard, and Fabien Alibart. Voltage-dependent synaptic plasticity: Unsupervised probabilistic hebbian plasticity rule based on neurons membrane potential. Frontiers in Neuroscience, 16:983950, 2022

    work page 2022

  9. [9]

    Unsupervised local learning based on voltage-dependent synaptic plasticity for resistive and ferroelectric synapses, Octo- ber 2025

    Nikhil Garg, Ismael Balafrej, Joao Henrique Quintino Palhares, Laura B´ egon-Lours, Davide Florini, Donato Francesco Falcone, Tommaso Stecconi, Valeria Bragaglia, Bert Jan Offrein, Jean-Michel Por- tal, Damien Querlioz, Yann Beilliard, Dominique Drouin, and Fabien Alibart. Unsupervised local learning based on voltage-dependent synaptic plasticity for resi...

    work page 2025

  10. [10]

    Unsupervised learning of digit recognition using spike-timing- dependent plasticity.Frontiers in computational neuroscience, 9:99, 2015

    Peter U Diehl and Matthew Cook. Unsupervised learning of digit recognition using spike-timing- dependent plasticity.Frontiers in computational neuroscience, 9:99, 2015

    work page 2015

  11. [11]

    Immunity to device varia- tions in a spiking neural network with memristive nanodevices.IEEE transactions on nanotechnology, 12(3):288–295, 2013

    Damien Querlioz, Olivier Bichler, Philippe Dollfus, and Christian Gamrat. Immunity to device varia- tions in a spiking neural network with memristive nanodevices.IEEE transactions on nanotechnology, 12(3):288–295, 2013

    work page 2013

  12. [12]

    Neuromorphic comput- ing with multi-memristive synapses.Nature communications, 9(1):2514, 2018

    Irem Boybat, Manuel Le Gallo, SR Nandakumar, Timoleon Moraitis, Thomas Parnell, Tomas Tuma, Bipin Rajendran, Yusuf Leblebici, Abu Sebastian, and Evangelos Eleftheriou. Neuromorphic comput- ing with multi-memristive synapses.Nature communications, 9(1):2514, 2018

    work page 2018

  13. [13]

    Unsupervised and efficient learning in sparsely activated convolutional spiking neural networks enabled by voltage-dependent synaptic plasticity

    Gaspard Goupy, Alexandre Juneau-Fecteau, Nikhil Garg, Ismael Balafrej, Fabien Alibart, Luc Frechette, Dominique Drouin, and Yann Beilliard. Unsupervised and efficient learning in sparsely activated convolutional spiking neural networks enabled by voltage-dependent synaptic plasticity. Neuromorphic Computing and Engineering, 3(1):014001, 2023. 13

    work page 2023