Pith. sign in

REVIEW 3 major objections 4 minor 81 references

Unveiling defect motifs in amorphous GeSe using machine learning interatomic potentials

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that the mid-gap defect states driving ovonic threshold switching in amorphous GeSe come from two chain-like motifs—aligned Ge chains near the conduction band and overcoordinated square-pyramidal Ge chains near the…

desk verdict A well-executed MLIP architecture study that identifies two plausible Ge-chain defect motifs in a-GeSe, but overreaches slightly in calling them the primary mid-gap defect centers without a direct DFT check on the production cells. read the letter →

arxiv 2506.15934 v1 pith:2BTGK36T submitted 2025-06-19 cond-mat.mtrl-sci

Minseok Moon , Seungwoo Hwang , Jaesun Kim , Yutack Park , Changho Hong , Seungwu Han This is my paper
classification cond-mat.mtrl-sci
keywords ovonicthresholdswitchingamorphousGeSemid-gapdefectstatesmachinelearninginteratomicpotentialgraphneuralnetworkchaindefectsPeierlsdistortionmelt-quenchmoleculardynamics
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

The paper sets out to identify the atomic configurations responsible for the localized gap states in amorphous GeSe, the material used in ovonic threshold switching (OTS) selectors. Using molecular dynamics with machine-learned interatomic potentials benchmarked against density functional theory, it finds that two chain-like motifs carry the defect physics: Ge chains with aligned bond angles host empty states near the conduction band, and Ge chains containing a five-fold coordinated, square-pyramidal Ge host occupied states near the valence band. It further shows that the defect levels are not fixed but vary continuously with the chain's average bond-angle alignment and the surrounding Peierls distortion, which would explain the broad, history-dependent trap distributions seen in experiments. A necessary prerequisite is that the potential captures at least four-body correlations and an extended effective interaction range; simpler potentials generate spurious cubic-like defects. If correct, the work gives a structural vocabulary for engineering OTS behavior through composition and processing.

What carries the argument

The argument is carried by an E(3)-equivariant graph neural network interatomic potential trained on density functional theory data, whose multi-layer message passing gives an effective receptive field of about 16 Å and whose self-interaction order captures four- and five-body correlations. The paper shows that three-body-limited potentials produce flat four-membered rings and artificial cubic-like motifs, while shallow networks with large cutoffs lack the depth to feel medium-range displacements. A structural taxonomy is built from 20 independent 960-atom melt-quenched cells, in which localized states are identified by inverse participation ratio and partial-charge isosurfaces and then assigned to atomic motifs. Two scalar descriptors of each motif carry the quantitative correlation: the average Ge–Ge–Ge bond angle along aligned chains, and the Peierls distortion ratio (the ratio of bond lengths between aligned bond pairs) around the overcoordinated Ge atom, both of which tune the defect energy position within the mobility gap.

What would settle it

If a high-resolution structural probe such as Ge K-edge extended X-ray absorption fine structure on melt-quenched amorphous GeSe finds no Ge atoms with a fifth Ge neighbor at distances consistent with the square-pyramidal motif, then the valence-side defect motif would not be the experimental trap center.

Watch

Extended reading notes

Core claim

The central claim is that the mid-gap defect states of amorphous GeSe come in two structural families, not from isolated point defects. Aligned Ge chains are strings of at least four Ge atoms with Ge–Ge–Ge bond angles above 120 degrees; they create unoccupied localized states that sit near the conduction band, and the deeper the state the larger the average bond angle. Overcoordinated Ge chains contain at least three Ge atoms with one Ge five-fold coordinated in a square-pyramidal arrangement; these create occupied localized states near the valence band, whose energy rises toward the band edge as the local Peierls distortion ratio grows. Strictly straight or strictly distorted chains that fall outside these geometric windows contribute only band-tail states, which is why earlier work found that not all Ge chains are defects. The paper also argues that the same motifs, plus reversible switching of non-defective chains into defect-active geometries, explain the broad deep-level transient spectroscopy peaks, the shift of band-tail states under bias seen in inverse photoemission, and the polarity-dependent threshold voltage behavior.

Load-bearing premise

The melt-quenched 960-atom cells produced by the graph neural network potential are assumed to faithfully represent the medium-range order and defect population of real amorphous GeSe, even though the potential was validated only against 120-atom DFT structures and structural statistics.

Editorial extensions

If this is right

  • Amorphous GeSe's trap spectrum is determined by medium-range chain topology, so engineering the glass (composition, quench rate, stoichiometry) should shift threshold voltages in a predictable direction.
  • The geometric correlations give a predictive rule: deepen conduction-side traps by increasing chain length and bond-angle alignment, and deepen valence-side traps by reducing the local Peierls distortion around five-fold Ge.
  • Machine-learning-potential-based large-cell sampling, when done with sufficient body order and receptive field, can replace the valence-alternation-point-defect picture for chalcogenide OTS materials.
  • Reversible transformation between band-tail chains and gap-state chains under bias offers a microscopic mechanism for the millisecond-timescale threshold voltage drift and for SET/RESET-dependent DLTS distributions.
  • The same method transfers to stoichiometric variations (GexSe1−x) and doped systems to map how dopants alter defect energies.

Reading between the lines

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

  • If the correlation is quantitative, a single experimental observable—such as the width of the DLTS peak—could be inverted to estimate the distribution of bond angles and Peierls ratios in the glass; the paper does not attempt this inversion.
  • The claim that three-body potentials faithfully represent medium-range order in other amorphous chalcogenides should be re-examined; the flat-ring artifact may contaminate earlier MLIP studies of Ge2Sb2Te5 and GeSe2.
  • A direct test would be to measure Ge K-edge EXAFS or fluctuation electron microscopy on melt-quenched GexSe1−x and look for signatures of four-or-more-atom Ge chains with bond angles above 120 degrees; if absent, the aligned-chain motif is an artifact of the finite-size or quench protocol.
  • The computed mobility gap of 0.7 eV is a density-functional-theory underestimate, so the absolute defect energy positions should be treated cautiously even if the qualitative band-edge assignment is correct.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper benchmarks three machine-learning interatomic potential architectures (BPNN, MTP, and SevenNet) on amorphous GeSe using a common training set, and argues that faithful modeling requires at least four-body correlations and an extended effective receptive field. The authors show that three-body-limited models produce artificial cubic motifs and flat four-membered rings, while shallow GNNs with large nominal cutoffs fail to suppress Ge-chain defects because of rapid sensitivity decay. Using a four-layer SevenNet model, they generate twenty 960-atom amorphous cells and identify two defect motifs: aligned Ge chains near the conduction band and overcoordinated Ge chains near the valence band. They further correlate defect energy levels with the average Ge-Ge-Ge bond angle and the Peierls distortion ratio, respectively, and discuss the implications for ovonic threshold switching experiments.

Significance. If correct, the paper provides a concrete, transferable design rule for MLIPs in amorphous chalcogenides—namely that four-body correlations and network depth, not just nominal cutoff, control medium-range structural fidelity—and proposes two specific, falsifiable structural motifs for OTS defect states in a-GeSe. The benchmarking is a strength: identical training data across architectures, multiple independent melt-quench runs, a quantitative sensitivity metric, and DFT single-point electronic-structure checks on MLIP-generated structures. The geometric correlations with defect energies are also specific enough to be tested by future DFT or experimental studies. The main weaknesses are the absence of direct DFT validation of the production-cell structural populations and the lack of statistical reporting for the defect-state correlations; these limit but do not invalidate the central claim.

major comments (3)
  1. [Section 3.2, Figs 9-11] The central two-motif claim rests on 20 960-atom cells generated by SevenNet, but the reported validation in this regime is limited to energy/force RMSE values (5.8 meV/atom, 0.1 eV/A) on snapshots. Even if these RMSE values are evaluated against DFT, they do not directly verify that the medium-range structural populations—aligned Ge chains, overcoordinated chains, and their lengths—match DFT. Because the training set contains only 120-atom melt-quench cells with 10 K/ps quenches, a bias toward short or specific chain configurations cannot be excluded by the reported metrics. Please add a DFT-based check of motif statistics, for example by comparing Ge-Ge-Ge angle distributions and chain-length distributions in 120-atom DFT cells against those in 960-atom MLIP cells, or by performing DFT relaxation and electronic-structure calculations on representative extracted motifs.
  2. [Section 3.2, Figs 10-11] The correlations between defect energy levels and geometric descriptors are presented without counts, correlation coefficients, or uncertainty estimates. The text states that seven-atom chains are outliers (stars) and that typical chains are four or five atoms, but the number of defect states contributing to each panel is not given, and no fraction of the total mid-gap states accounted for by the two motifs is reported. Without these numbers, the reader cannot judge whether the trends are statistically robust or whether the two-motif classification is exhaustive. Please report the number of chains/states per motif, R-squared or equivalent, and a completeness statement for all localized gap states identified in the 20 cells.
  3. [Section 3.2, motif definitions] The motif definitions are introduced post hoc and depend on thresholds (at least four Ge atoms, Ge-Ge-Ge angles above 120 degrees, at least three Ge atoms with a five-fold coordinated square-pyramidal Ge). No sensitivity analysis is provided for these thresholds, and there is no systematic search for alternative structural descriptors. Because these thresholds determine which states are classified as defect motifs, their arbitrariness weakens the claim that exactly two motifs exist. Please provide a sensitivity analysis or apply an unsupervised structural classifier to demonstrate that the two-motif classification is not an artifact of the chosen thresholds.
minor comments (4)
  1. [Section 3.1.3] The text says 'Representative DOS and IPR results for each variant are compared in Figure 5', but the corresponding panels are in Figure 7; Figure 5 already shows the cubicity distributions.
  2. [Page 16, near Ref. 69] The phrase 'reveal distinct distinct distributions of defect levels' contains a duplicated word; remove the second 'distinct'.
  3. [Figure 9 caption] The symbol X^IV in the schematics is not defined in the caption or the surrounding text; please define it explicitly, noting that it denotes a four-coordinated Ge or Se atom.
  4. [Generally] The manuscript would benefit from a data and code availability statement, since the reproducibility of the MLIP training and the defect analysis depends on the exact training sets, hyperparameters, and analysis scripts.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: defect motifs are identified from DFT-computed electronic states on MLIP-generated structures, and the bond-angle/Peierls correlations are empirical, not fitted.

full rationale

The central derivation chain is self-contained against DFT benchmarks. The MLIPs are trained on DFT AIMD data and validated against independent DFT-derived 120-atom amorphous cells via RDF/ADF, ring statistics, DOS, and IPR. The 960-atom production cells are generated with SevenNet, but the electronic defect levels are computed by DFT (PBE, Gamma-point) on those geometries, not by the MLIP. The two defect motifs are obtained by locating localized mid-gap states in the DFT DOS/IPR and then finding common structural environments, so the motifs are not defined in terms of the predicted energies. The correlations in Figures 10-11 plot geometrically defined descriptors (average Ge-Ge-Ge bond angle, Peierls distortion ratio) against DFT defect energies; no parameter is fitted to those energies, so the trends are not forced by construction. The stated 960-atom RMSE (5.8 meV/atom, 0.1 eV/A) is a prediction-error check on MLIP-generated snapshots; although the sentence does not explicitly say 'against DFT', the DOS on the same cells is explicitly DFT and the values are comparable to the DFT test-set errors, so this is an external consistency check rather than a self-fit. Self-citations to SevenNet and SIMPLE-NN are tool/package citations and are not load-bearing: the architecture conclusions (four-body correlation order and network-depth-dependent receptive field) are demonstrated in-paper by comparison to DFT structures. The residual concern that the 960-atom cells may not faithfully represent experimental medium-range order, and that experimental validation of the motifs is still needed, is a correctness/generalizability risk, not circularity.

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

The central claim is not derived from a fitted parameter, but it does rest on hand-chosen classification thresholds and on DFT/MD approximations. The MLIP weights themselves are fitted to DFT training data, which is standard practice and not circular.

free parameters (4)
  • Aligned Ge chain length and angle thresholds = >= 4 Ge atoms and > 120 degrees
    Defined in Section 3.2 to classify aligned Ge chains as a defect motif; the correlation in Figure 10 depends on this hand-picked threshold.
  • Overcoordinated Ge chain thresholds = >= 3 Ge atoms with at least one five-fold coordinated Ge
    Defined in Section 3.2 to classify overcoordinated Ge chains; the valence-band defect assignment depends on this criterion.
  • Peierls distortion ratio cutoff for gap versus tail states = approximately 1.15
    Section 3.2 states that overcoordinated chains with distortion ratios above ~1.15 contribute to tail states rather than gap states, separating the two regimes.
  • Mobility edges from DOS fitting = E_c and E_v yielding a mobility gap of ~0.7 eV
    Section 2.3 fits the DOS to (E_v - E)^{1/2} and (E - E_c)^{1/2}; the classification of defect states as inside or outside the gap depends on these fitted edges.
assumptions (4)
  • domain assumption PBE functional describes defect-state energetics in a-GeSe sufficiently for identifying mid-gap states
    Invoked in Section 2.3; the paper acknowledges band gap underestimation but assumes PBE reproduces the electronic structure of GeSe (Ref. 56) and uses PBE for all defect analysis.
  • domain assumption Melt-quench MD at 10 K/ps on 960-atom cells produces amorphous GeSe representative of experimental a-GeSe
    Used throughout Section 3.2; no direct experimental structural validation of the 960-atom cells is given, only MLIP-vs-DFT consistency on 120-atom cells.
  • domain assumption Gamma-point sampling is sufficient for DOS/IPR of 960-atom amorphous cells
    Section 2.3 uses Gamma-point for 960-atom cells; the paper does not test k-point convergence for defect levels.
  • domain assumption Defect states are correctly identified as localized states within the fitted mobility gap
    The paper defines defect states using mobility edges fitted to a Tauc-like DOS form (Section 2.3); states slightly outside the fitted gap are classified as tail states.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unveiling defect motifs in amorphous GeSe using machine learning interatomic potentials." pith.science (2026). https://pith.science/paper/2BTGK36T

@misc{pith2026250615934,
  author       = {Pith},
  title        = {Pith review of: Unveiling defect motifs in amorphous GeSe using machine learning interatomic potentials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2BTGK36T}},
  note         = {Machine review of arXiv:2506.15934}
}
read the original abstract

Ovonic threshold switching (OTS) selectors play a critical role in non-volatile memory devices because of their nonlinear electrical behavior and polarity-dependent threshold voltages. However, the atomic-scale origins of the defect states responsible for these properties are not yet fully understood. In this study, we use molecular dynamics simulations accelerated by machine-learning interatomic potentials to investigate defects in amorphous GeSe. We begin by benchmarking several potential architectures-including descriptor-based models and graph neural network (GNN) models-and show that faithfully representing amorphous GeSe requires capturing higher-order interactions (at least four-body correlations) and medium-range structural order. We find that GNN architectures with multiple interaction layers successfully capture these correlations and structural motifs, preventing the spurious defects that less expressive models introduce. With our optimized GNN potential, we examine twenty independent 960-atom amorphous GeSe structures and identify two distinct defect motifs: aligned Ge chains, which give rise to defect states near the conduction band, and overcoordinated Ge chains, which produce defect states near the valence band. We further correlate these electronic defect levels with specific structural features-namely, the average alignment of bond angles in the aligned chains and the degree of local Peierls distortion around overcoordinated Ge atoms. These findings provide a theoretical framework for interpreting experimental observations and deepen our understanding of defect-driven OTS phenomena in amorphous GeSe.

Figures

Figures reproduced from arXiv: 2506.15934 by the authors.

Figure 1
Figure 1. Electronic DOS (black lines) and IPR (red dots) for amorphous structures generated [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Partial charge distributions illustrat￾ing localized mid-gap states associated with (a) cubic-like motifs from the BPNN model and (b) aligned Ge–Ge chains from the MTP model. Charge density isosurfaces highlight regions of localization. Only atoms contributing to the lo￾calized mid-gap charge density are shown. average deviations are defined as: ϵℓ = 1 12 P12 i=1|ℓi − ¯ℓ| ¯ℓ , ϵθ = 1 24 P24 i=1|θi − 90| 90 (2) where… view at source ↗
Figure 3
Figure 3. Atomic configurations of cubic struc￾tures extracted from amorphous GeSe models with (a) low and (b) high cubicity (C) [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Histogram of cubicity distributions for cubic motifs identified in a-GeSe produced by DFT and various descriptor-type MLIPs. and DFT structures. Furthermore, the MTP￾low model generates localized defects centered around these high-cubicity motifs, closely re￾sembling t…
Figure 5
Figure 5. Figure 5: Histogram of cubicity distribu￾tions for cubic motifs identified in amorphous GeSe structures generated by DFT and single￾interaction-layer SevenNet models with varying maximum correlation order (ν). To further explore this behavior, we ana￾lyze ring populations using …
Figure 6
Figure 6. Figure 6: Comparison of square-ring distributions and ring geometry between DFT and MLIP [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Electronic DOS (black lines) and IPR (red dots) of amorphous structures generated using [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Distance dependence of the normal [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Schematics, partial charge distributions, and defect energy levels illustrating two dis [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 11
Figure 11. Figure 11: Scatter plot illustrating defect en [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

81 extracted references · 63 canonical work pages

  1. [1]

    A review of emerging non-volatile memory (NVM) technologies and applications

    Chen, A. A review of emerging non-volatile memory (NVM) technologies and applications. Solid-State Electronics 2016, 125, 25--38

    2016

  2. [2]

    Ovonic threshold switching selectors for three-dimensional stackable phase-change memory

    Zhu, M.; Ren, K.; Song, Z. Ovonic threshold switching selectors for three-dimensional stackable phase-change memory . MRS Bull. 2019, 44, 715--720

    work page 2019

  3. [3]

    Phase change materials and phase change memory

    Raoux, S.; Xiong, F.; Wuttig, M.; Pop, E. Phase change materials and phase change memory. MRS bulletin 2014, 39, 703--710

    work page 2014

  4. [4]

    L.; Kundu, S.; Hody, H.; Devulder, W.; Houdt, J

    Ravsher, T.; Garbin, D.; Fantini, A.; Degraeve, R.; Clima, S.; Donadio, G. L.; Kundu, S.; Hody, H.; Devulder, W.; Houdt, J. V.; Afanas'ev, V.; Delhougne, R.; Kar, G. S. Polarity‐induced threshold voltage shift in ovonic threshold switching chalcogenides and the impact of material composition . Phys. Status Solidi RRL 2023, 17

    work page 2023

  5. [5]

    L.; Kundu, S.; Hody, H.; Devulder, W.; Houdt, J

    Ravsher, T.; Garbin, D.; Fantini, A.; Degraeve, R.; Clima, S.; Donadio, G. L.; Kundu, S.; Hody, H.; Devulder, W.; Houdt, J. V.; Afanas’ev, V.; Delhougne, R.; Kar, G. S. Self-rectifying memory cell based on SiGeAsSe ovonic threshold switch . IEEE Trans. Electron Devices 2023, 70, 2276--2281

    work page 2023

  6. [6]

    Clima, S. et al. Ovonic threshold-switching Ge_ x Se_ y chalcogenide materials: stoichiometry, trap nature, and material relaxation from first principles. Phys. Status Solidi RRL 2020, 14, 1900672

    work page 2020

  7. [7]

    Chalcogenide ovonic threshold switching selector

    Zhao, Z.; Clima, S.; Garbin, D.; Degraeve, R.; Pourtois, G.; Song, Z.; Zhu, M. Chalcogenide ovonic threshold switching selector . Nano-Micro Lett. 2024, 16, 81

    work page 2024

  8. [8]

    Phase-change materials for non-volatile memory devices: from technological challenges to materials science issues

    Noé, P.; Vallée, C.; Hippert, F.; Fillot, F.; Raty, J.-Y. Phase-change materials for non-volatile memory devices: from technological challenges to materials science issues . Semicond. Sci. Technol. 2017, 33, 013002

    work page 2017

Show all 81 references
  1. [9]

    Govoreanu, B. et al. Thermally stable integrated Se-based OTS selectors with >20 MA/cm^2 current drive, >3.10^3 half-bias nonlinearity, tunable threshold voltage and excellent endurance . Symposium on VLSI Technology 2017, T92--T93

    2017

  2. [10]

    L.; Witters, T.; Kundu, S.; Goux, L.; Afanasiev, V.; Kar, G.; Pourtois, G

    Clima, S.; Govoreanu, B.; Opsomer, K.; Velea, A.; Avasarala, N.; Devulder, W.; Shlyakhov, I.; Donadio, G. L.; Witters, T.; Kundu, S.; Goux, L.; Afanasiev, V.; Kar, G.; Pourtois, G. Atomistic investigation of the electronic structure, thermal properties and conduction defects i...

    2017

  3. [11]

    GeSe ovonic threshold switch: the impact of functional layer thickness and device size

    Zhao, J.; Zhao, Z.; Song, Z.; Zhu, M. GeSe ovonic threshold switch: the impact of functional layer thickness and device size . Sci. Rep. 2024, 14, 6685

    2024

  4. [12]

    S.; Lim, H.; Park, G.-H.; Hwang, C

    Jeong, D. S.; Lim, H.; Park, G.-H.; Hwang, C. S.; Lee, S.; Cheong, B.-k. Threshold resistive and capacitive switching behavior in binary amorphous GeSe . J. Appl. Phys. 2012, 111, 102807

    2012

  5. [13]

    Extended endurance performance and reduced threshold voltage by doping Si in GeSe-based ovonic threshold switching selectors

    Li, X.; Yuan, Z.; Lv, S.; Song, S.; Song, Z. Extended endurance performance and reduced threshold voltage by doping Si in GeSe-based ovonic threshold switching selectors . Thin Solid Films 2021, 734, 138837

    2021

  6. [14]

    S.; Detavernier, C

    Keukelier, J.; Opsomer, K.; Devulder, W.; Clima, S.; Goux, L.; Kar, G. S.; Detavernier, C. Tuning of the thermal stability and ovonic threshold switching properties of GeSe with metallic and non-metallic alloying elements . J. Appl. Phys. 2021, 130, 165103

    2021

  7. [15]

    V.; Karpov, V

    Nardone, M.; Simon, M.; Karpov, I. V.; Karpov, V. G. Electrical conduction in chalcogenide glasses of phase change memory . J. Appl. Phys. 2012, 112, 071101

    2012

  8. [16]

    Analytical model for subthreshold conduction and threshold switching in chalcogenide-based memory devices

    Ielmini, D.; Zhang, Y. Analytical model for subthreshold conduction and threshold switching in chalcogenide-based memory devices. J. Appl. Phys. 2007, 102, 054517

    2007

  9. [17]

    A HydroDynamic model for trap-assisted tunneling conduction in ovonic devices

    Buscemi, F.; Piccinini, E.; Vandelli, L.; Nardi, F.; Padovani, A.; Kaczer, B.; Garbin, D.; Clima, S.; Degraeve, R.; Kar, G. A HydroDynamic model for trap-assisted tunneling conduction in ovonic devices. IEEE Trans. Electron Devices 2023, 70, 1808--1814

    2023

  10. [18]

    Current-driven threshold switching of a small polaron semiconductor to a metastable conductor

    Emin, D. Current-driven threshold switching of a small polaron semiconductor to a metastable conductor. Phys. Rev. B 2006, 74, 035206

    2006

  11. [19]

    Degraeve, R.; Ravsher, T.; Kabuyanagi, S.; Fantini, A.; Clima, S.; Garbin, D.; Kar, G. S. Modeling and spectroscopy of ovonic threshold switching defects. IEEE International Reliability Physics Symposium 2021, 1--5

    2021

  12. [20]

    A unified model of nucleation switching

    Nardone, M.; Karpov, V.; Jackson, D.; Karpov, I. A unified model of nucleation switching. Appl. Phys. Lett. 2009, 94

    2009

  13. [21]

    Amorphous chalcogenide semiconductors and related materials, 2nd ed.; Springer Cham, 2021

    Tanaka, K.; Shimakawa, K. Amorphous chalcogenide semiconductors and related materials, 2nd ed.; Springer Cham, 2021

    2021

  14. [22]

    G.; Strom, U.; Taylor, P

    Bishop, S. G.; Strom, U.; Taylor, P. C. Optically induced metastable paramagnetic states in amorphous semiconductors . Phys. Rev. B 1977, 15, 2278--2294

    1977

  15. [23]

    Low temperature photoluminescence and fatigue effects in Se Ge glasses

    Ball, G.; Chamberlain, J.; Instone, T. Low temperature photoluminescence and fatigue effects in Se Ge glasses . Solid State Commun. 1978, 27, 71--74

    1978

  16. [24]

    A.; Koóos, M.; Somogyi, I

    Vassilyev, V. A.; Koóos, M.; Somogyi, I. Radiative localized centres in Ge-Se glasses observed by photoluminescence . Philos. Mag. B 1979, 39, 333--348

    1979

  17. [25]

    Mollot, F.; Cernogora, J.; Guillaume, C. B. Excitation spectra of photoluminescence fatigue and creation of paramagnetic centers in amorphous Ge_ x Se_ 1-x . J. Non-Cryst. Solids 1980, 35, 939--944

    1980

  18. [26]

    I.; Inuishi, Y

    Shirafuji, J.; Kim, G. I.; Inuishi, Y. Photoluminescence and optical properties of Ge_ 1-x Se_ x glasses . Jpn. J. Appl. Phys. 1977, 16, 67

    1977

  19. [27]

    Valence-alternation model for localized gap states in lone-pair semiconductors

    Kastner, M.; Adler, D.; Fritzsche, H. Valence-alternation model for localized gap states in lone-pair semiconductors . Phys. Rev. Lett. 1976, 37, 1504--1507

    1976

  20. [28]

    S.; Silver, M.; Ovshinsky, S

    Adler, D.; Shur, M. S.; Silver, M.; Ovshinsky, S. R. Threshold switching in chalcogenide-glass thin films . J. Appl. Phys. 1980, 51, 3289--3309

    1980

  21. [29]

    Deep machine learning unravels the structural origin of mid-gap states in chalcogenide glass for high-density memory integration

    Xu, M.; Xu, M.; Miao, X. Deep machine learning unravels the structural origin of mid-gap states in chalcogenide glass for high-density memory integration. InfoMat 2022, 4, e12315

    2022

  22. [30]

    Material relaxation in chalcogenide OTS selector materials

    Clima, S.; Garbin, D.; Devulder, W.; Keukelier, J.; Opsomer, K.; Goux, L.; Kar, G.; Pourtois, G. Material relaxation in chalcogenide OTS selector materials . Microelectron. Eng. 2019, 215, 110996

    2019

  23. [31]

    Sb‐Se‐based electrical switching device with fast transition speed and minimized performance degradation due to stable mid‐gap states

    Mai, X.; Xu, Q.; Yang, Z.; Wang, H.; Liu, Y.; Shen, Y.; Hu, H.; Xu, M.; Wang, Z.; Tong, H.; Wang, C.; Miao, X.; Xu, M. Sb‐Se‐based electrical switching device with fast transition speed and minimized performance degradation due to stable mid‐gap states . Electron 2024,

    2024

  24. [32]

    Device‐to‐materials pathway for electron traps detection in amorphous GeSe‐based selectors

    Slassi, A.; Medondjio, L.; Padovani, A.; Tavanti, F.; He, X.; Clima, S.; Garbin, D.; Kaczer, B.; Larcher, L.; Ordejón, P.; Calzolari, A. Device‐to‐materials pathway for electron traps detection in amorphous GeSe‐based selectors . Adv. Electron. Mater. 2023, 9

    2023

  25. [33]

    Crystallization of amorphous GeTe simulated by neural network potential addressing medium-range order

    Lee, D.; Lee, K.; Yoo, D.; Jeong, W.; Han, S. Crystallization of amorphous GeTe simulated by neural network potential addressing medium-range order. Comput. Mater. Sci. 2020, 181, 109725

    2020

  26. [34]

    C.; Deringer, V

    Sosso, G. C.; Deringer, V. L.; Elliott, S. R.; Cs \'a nyi, G. Understanding the thermal properties of amorphous solids using machine-learning-based interatomic potentials. Mol. Simul. 2018, 44, 866--880

    2018

  27. [35]

    C.; Bernasconi, M

    Sosso, G. C.; Bernasconi, M. Harnessing machine learning potentials to understand the functional properties of phase-change materials. MRS Bull. 2019, 44, 705--709

    2019

  28. [36]

    C.; Konstantinou, K.; Lee, T

    Mocanu, F. C.; Konstantinou, K.; Lee, T. H.; Bernstein, N.; Deringer, V. L.; Cs \'a nyi, G.; Elliott, S. R. Modeling the phase-change memory material, Ge_ 2 Sb_ 2 Te_ 5 , with a machine-learned interatomic potential. J. Phys. Chem. B 2018, 122, 8998--9006

    2018

  29. [37]

    Quench-rate and size-dependent behaviour in glassy Ge_ 2 Sb_ 2 Te_ 5 models simulated with a machine-learned Gaussian approximation potential

    Mocanu, F.; Konstantinou, K.; Elliott, S. Quench-rate and size-dependent behaviour in glassy Ge_ 2 Sb_ 2 Te_ 5 models simulated with a machine-learned Gaussian approximation potential. J. Phys. D: Appl. Phys. 2020, 53, 244002

    2020

  30. [38]

    L.; Zhang, W

    Zhou, Y.-X.; Zhang, H.-Y.; Deringer, V. L.; Zhang, W. Structure and Dynamics of Supercooled Liquid Ge_ 2 Sb_ 3 Te_ 2 from Machine-Learning-Driven Simulations. Phys. Status Solidi RRL 2021, 15, 2000403

    2021

  31. [39]

    P.; Shimamura, K.; Fukushima, S.; Shimojo, F.; Kalia, R.; Nakano, A.; Vashishta, P

    Rajak, P.; Baradwaj, N.; Nomura, K.-i.; Krishnamoorthy, A.; Rino, J. P.; Shimamura, K.; Fukushima, S.; Shimojo, F.; Kalia, R.; Nakano, A.; Vashishta, P. Neural network quantum molecular dynamics, intermediate range order in GeSe_ 2 , and neutron scattering experiments . J. Phy...

    2021

  32. [40]

    Generalized neural-network representation of high-dimensional potential-energy surfaces

    Behler, J.; Parrinello, M. Generalized neural-network representation of high-dimensional potential-energy surfaces. Phys. Rev. Lett. 2007, 98, 146401

    2007

  33. [41]

    Shapeev, A. V. Moment tensor potentials: a class of systematically improvable interatomic potentials. Multiscale Modeling & Simulation 2016, 14, 1153--1173

    2016

  34. [42]

    Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations

    Park, Y.; Kim, J.; Hwang, S.; Han, S. Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations. J. Chem. Theory Comput. 2024, 20, 4857--4868

    2024

  35. [43]

    A generalizable, uncertainty-aware neural network potential for GeSbTe with Monte Carlo dropout

    Lee, S.-H.; Olevano, V.; Skl \'e nard, B. A generalizable, uncertainty-aware neural network potential for GeSbTe with Monte Carlo dropout. Solid-State Electronics 2023, 199, 108508

    2023

  36. [44]

    M.; Jang, J.; Yoon, S.; Ghosh, A.; Kim, M.; Kim, J.; Na, W.; Kim, J.; Choi, H

    Lee, S.; Jung, J.-E.; Kim, H.-g.; Lee, Y.; Park, J. M.; Jang, J.; Yoon, S.; Ghosh, A.; Kim, M.; Kim, J.; Na, W.; Kim, J.; Choi, H. J.; Cheong, H.; Kim, K. -GeSe: a new hexagonal polymorph from group IV--VI monochalcogenides. Nano Lett. 2021, 21, 4305--4313

    2021

  37. [45]

    Change of short-range order with temperature and composition in liquid Ge_ x Se_ 1-x as shown by density measurements

    Ruska, J.; Thurn, H. Change of short-range order with temperature and composition in liquid Ge_ x Se_ 1-x as shown by density measurements. J. Non-Cryst. Solids 1976, 22, 277--290

    1976

  38. [46]

    Atomic energy mapping of neural network potential

    Yoo, D.; Lee, K.; Jeong, W.; Lee, D.; Watanabe, S.; Han, S. Atomic energy mapping of neural network potential. Phys. Rev. Mater. 2019, 3, 093802

    2019

  39. [47]

    P.; Kornbluth, M.; Molinari, N.; Smidt, T

    Batzner, S.; Musaelian, A.; Sun, L.; Geiger, M.; Mailoa, J. P.; Kornbluth, M.; Molinari, N.; Smidt, T. E.; Kozinsky, B. E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials. Nat. Commun. 2022, 13, 2453

    2022

  40. [48]

    P.; Simm, G.; Ortner, C.; Cs \'a nyi, G

    Batatia, I.; Kovacs, D. P.; Simm, G.; Ortner, C.; Cs \'a nyi, G. MACE: higher order equivariant message passing neural networks for fast and accurate force fields. Adv. Neural. Inf. Process. Syst. 2022, 35, 11423--11436

    2022

  41. [49]

    SIMPLE-NN: an efficient package for training and executing neural-network interatomic potentials

    Lee, K.; Yoo, D.; Jeong, W.; Han, S. SIMPLE-NN: an efficient package for training and executing neural-network interatomic potentials. Computer Phys. Commun. 2019, 242, 95--103

    2019

  42. [50]

    Atom-centered symmetry functions for constructing high-dimensional neural network potentials

    Behler, J. Atom-centered symmetry functions for constructing high-dimensional neural network potentials. J. Chem. Phys. 2011, 134

    2011

  43. [51]

    Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set

    Kresse, G.; Furthm \"u ller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 1996, 54, 11169

    1996

  44. [52]

    Bl \"o chl, P. E. Projector augmented-wave method. Phys. Rev. B 1994, 50, 17953

    1994

  45. [53]

    P.; Burke, K.; Ernzerhof, M

    Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 1996, 77, 3865

    1996

  46. [54]

    Mean-value point in the Brillouin zone

    Baldereschi, A. Mean-value point in the Brillouin zone. Phys. Rev. B 1973, 7, 5212

    1973

  47. [55]

    P.; Levy, M

    Perdew, J. P.; Levy, M. Physical content of the exact Kohn-Sham orbital energies: band gaps and derivative discontinuities. Phys. Rev. Lett. 1983, 51, 1884

    1983

  48. [56]

    Enhanced photocatalytic activity for water splitting of blue-phase GeS and GeSe monolayers via biaxial straining

    Gu, D.; Tao, X.; Chen, H.; Zhu, W.; Ouyang, Y.; Peng, Q. Enhanced photocatalytic activity for water splitting of blue-phase GeS and GeSe monolayers via biaxial straining. Nanoscale 2019, 11, 2335--2342

    2019

  49. [57]

    Atomistic structure of band-tail states in amorphous silicon

    Dong, J.; Drabold, D. Atomistic structure of band-tail states in amorphous silicon. Phys. Rev. Lett. 1998, 80, 1928

    1998

  50. [58]

    T.; Frost, J

    Butler, K. T.; Frost, J. M.; Walsh, A. Band alignment of the hybrid halide perovskites CH3NH3PbCl3 , CH3NH3PbBr3 , and CH3NH3PbI3 . Mater. Horiz. 2015, 2, 228--231

    2015

  51. [59]

    L.; Wei, S.-H

    Walsh, A.; Da Silva, J. L.; Wei, S.-H. Multi-component transparent conducting oxides: progress in materials modelling. J. Condens. Matter. Phys. 2011, 23, 334210

    2011

  52. [60]

    Amorphous and liquid semiconductors; Springer: New York, 1974; pp 159--220

    Tauc, J. Amorphous and liquid semiconductors; Springer: New York, 1974; pp 159--220

    1974

  53. [61]

    P.; Musaelian, A.; Simm, G

    Batatia, I.; Batzner, S.; Kov \'a cs, D. P.; Musaelian, A.; Simm, G. N.; Drautz, R.; Ortner, C.; Kozinsky, B.; Cs \'a nyi, G. The design space of E(3)-equivariant atom-centred interatomic potentials. Nat. Mach. Intell. 2025, 7, 56--67

    2025

  54. [62]

    Ring statistics analysis of topological networks: new approach and application to amorphous GeS_ 2 and SiO_ 2 systems

    Le Roux, S.; Jund, P. Ring statistics analysis of topological networks: new approach and application to amorphous GeS_ 2 and SiO_ 2 systems. Comput. Mater. Sci. 2010, 49, 70--83

    2010

  55. [63]

    P.; Dong, X.; Bronstein, M

    Topping, J.; Di Giovanni, F.; Chamberlain, B. P.; Dong, X.; Bronstein, M. M. Understanding over-squashing and bottlenecks on graphs via curvature. arXiv preprint arXiv:2111.14522 2021,

    2021 arXiv

  56. [64]

    Optical and electrical properties of GeSe and SnSe single crystals

    Kim, Y.; Choi, I.-H. Optical and electrical properties of GeSe and SnSe single crystals. J. Korean Phys. Soc. 2018, 72, 238--242

    2018

  57. [65]

    M.; Biacchi, A

    Vaughn, D.; Sun, D.; Levin, S. M.; Biacchi, A. J.; Mayer, T. S.; Schaak, R. E. Colloidal synthesis and electrical properties of GeSe nanobelts. Chem. Mater. 2012, 24, 3643--3649

    2012

  58. [66]

    Hsu, P.-C.; Simoen, E.; Lin, D.; Stesmans, A.; Goux, L.; Delhougne, R.; Carolan, P.; Bender, H.; Kar, G. S. A deep level transient spectroscopy study of hole traps in Ge_ x Se_ 1-x -based Layers for ovonic threshold switching selectors. ECS J. Solid State Sci. Technol. 2020, 9, 044006

    2020

  59. [67]

    Controllable threshold voltage (Vth) drift in ovonic threshold switch devices under a high-frequency continuous operation

    Chen, Z.; Wang, L.; Wen, J.; Tong, H.; Miao, X. Controllable threshold voltage (Vth) drift in ovonic threshold switch devices under a high-frequency continuous operation . IEEE Trans. Electron Devices 2022, 69, 3158--3162

    2022

  60. [68]

    Clima, S. et al. Ovonic threshold‐switching Ge_ x Se_ y chalcogenide materials: stoichiometry, trap Nature, and material relaxation from first principles . Phys. Status Solidi RRL 2020, 14

    2020

  61. [69]

    Sung, H.-J.; Choi, M.; Wu, Z.; Chae, H.; Heo, S.; Kang, Y.; Koo, B.; Park, J.-B.; Yang, W.; Park, Y.; Ham, Y.; Kiyeon, Y.; Lee, C. S. Microscopic origin of polarity-dependent Vth shift in amorphous chalcogenides for 3D self-selecting memory. Adv. Sci. 2024, 11, 2408028

    2024

  62. [70]

    u tt, Kristof T and Sauceda, Huziel E and Kindermans, P-J and Tkatchenko, Alexandre and M \

    Seong, Y.-W.; Kwon, H.; Lee, C.; Lim, H.; Jeong, K.; Shin, H. J.; Cho, M.-H. Transient structural transition in ovonic threshold switching glass. Adv. Funct. Mater. 2025, 35, 2415462 mcitethebibliography main.tex0000664000000000000000000022004615024425464011236 0ustar rootroot...

    2025

  63. [71]

    write newline

    " write newline " cite write " FUNCTION editor.postfix editor num.names #1 > "( )" "( )" if FUNCTION editor.trans.postfix editor num.names #1 > "( )" "( )" if FUNCTION trans.postfix translator num.names #1 > "( )" "( )" if FUNCTION authors.editors.reflist.apa5 'field := 'dot :...

  64. [72]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year eprint archive archivePrefix primaryClass adsurl adsnote version label INTEGERS o...

  65. [73]

    write newline

    " write newline "" before.all 'output.state := FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap duplicate "7" = swap duplicate "8" = swap "9" = or or or or or or...

  66. [74]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter doi edition editor eid eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url volume year archivePrefix primaryClass adsurl adsnote version label extra.labe...

  67. [75]

    write newline

    " write newline "" before.all 'output.state := FUNCTION add.period duplicate empty 'skip "." * add.blank if FUNCTION if.digit duplicate "0" = swap duplicate "1" = swap duplicate "2" = swap duplicate "3" = swap duplicate "4" = swap duplicate "5" = swap duplicate "6" = swap dupl...

  68. [76]

    write newline

    " write newline "" before.all 'output.state := FUNCTION output.doi doi empty skip "doi:" doi * "" * output if FUNCTION format.archive archivePrefix empty "" archivePrefix ":" * if FUNCTION format.primaryClass primaryClass empty "" " [" primaryClass * "] " * if FUNCTION format....

  69. [77]

    write newline

    " write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

  70. [78]

    , " * write output.state after.block = add.period write newline

    ENTRY address archive author booktitle chapter edition editor eprint howpublished institution journal key keywords month note number organization pages publisher school series title type url doi volume year archivePrefix primaryClass eid adsurl adsnote version label INTEGERS o...

  71. [79]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  72. [80]

    Available from:

    ENTRY address assignee author booktitle chapter cartographer day edition editor howpublished institution inventor journal key keywords month note number organization pages part publisher school series title type volume word year eprint doi url lastchecked updated archive archi...

  73. [81]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.