Pith. sign in

REVIEW 4 major objections 5 minor 4 references

Size-Dependent Band-Tail Localization in Oxide Semiconductors Revealed by Direct Density-of-States Mapping

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read By directly mapping the density of states in working oxide transistors, this paper shows that the transport-active band-tail states shrink exponentially with channel length—a signature of disorder-induced localization that shifts the…

desk verdict Plausible and potentially useful new measurement campaign, but the central exponential-length claim rests on an untested chemical-potential alignment and error-bar-free fits, so it needs revision before the localization interpretation is solid. read the letter →

arxiv 2608.09283 v1 pith:OAHO7PR7 submitted 2026-08-10 cond-mat.mes-hall cond-mat.mtrl-sciphysics.app-ph

Chang Niu , Kisoo Nam , Aravindh Shankar , Jian-Yu Lin , Sanjeev Khare , Sumi Lee , Pramey Upadhyaya , Peide D. Ye This is my paper · ORCID
classification cond-mat.mes-hallcond-mat.mtrl-sciphysics.app-ph
keywords oxidesemiconductorsthin-filmtransistorsdensity-of-statesmappingband-tailstateslocalizationlengthquantumcapacitanceindiumdisorderengineering
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

Thin-film oxide transistors are usually described as if their band-tail states were fixed traps that only affect how well the gate switches the device. This paper argues that, in ultrathin In-based oxide channels, a large part of the band tail is instead made of states that become localized as the device gets longer, so the number of states that can actually screen and carry current shrinks exponentially with channel length. The evidence comes from a lock-in electric-field penetration technique that directly maps the density of states in working dual-gate transistors, from depletion to accumulation. If the claims hold, the effective density of states is not a material constant: it depends on device geometry through a localization length, and this size dependence is a new microscopic mechanism behind threshold-voltage roll-off in scaled oxide transistors.

What carries the argument

The central object is the effective density of states $D_{\mathrm{eff}}(E,L)$ measured by a dual-gate lock-in technique: an AC back-gate excitation penetrates the thin semiconductor channel and the normalized top-gate current gives the quantum capacitance $C_q = e^2\,\mathrm{d}n_s/\mathrm{d}\mu = e^2 D(E)$, from which carrier density and chemical potential are integrated. The load-bearing identity is the exponential suppression $D_{\mathrm{eff}}(E,L) = D_0(E)\exp[-L/\xi(E)]$, where $\xi(E)$ is the energy-dependent localization length; the slope of the $\ln D_{\mathrm{eff}}$ versus $L$ plot at fixed energy yields $1/\xi(E)$. This converts the abstract idea of Anderson localization into a measurable device-level scaling law.

What would settle it

Take the longest and shortest devices from the same film, measure their intrinsic depletion-onset voltages (where the penetration current plateaus), and re-align the chemical-potential axes using those onsets instead of the zero-bias reference; if the exponential suppression of $D_{\mathrm{eff}}$ with $L$ survives the re-alignment, the localization interpretation is supported, and if it collapses, the effect is an artifact of misaligned energy scales.

Watch

Extended reading notes

Core claim

Using the electric-field penetration technique, the authors extract the quantum capacitance, carrier density, and chemical potential of InOx and IGZO thin-film transistors and resolve four energy regimes: deep trap states, shallow band tails, disorder-dominated tail states, and extended diffusive states. The central quantitative result is that the effective transport-active DOS at fixed energy decays exponentially with channel length, $D_{\mathrm{eff}}(E,L) = D_0(E)\,\exp[-L/\xi(E)]$, while an independent localization length from conductance scaling falls in the same range. Temperature serves as the cross-check: the disorder-dominated portion of the tail is strongly suppressed at low temperature, whereas the extended states are left nearly unchanged. The same geometry and temperature signatures are reduced by thickening the film, by O$_2$ annealing, and by lowering In content through Ga/Zn substitution, which the authors read as systematic suppression of disorder. Together these observations are used to claim that the threshold-voltage roll-off seen at longer channels in these oxides is a localization effect rather than a contact or static-trap effect.

Load-bearing premise

The entire length-dependent comparison assumes that every device on the same film has the same chemical potential at zero back-gate bias, so that a fixed gate voltage means the same energy in every channel; if individual devices deplete at different voltages, the reported exponential decay could be an artifact of misaligned energy scales.

Editorial extensions

If this is right

  • Compact models and scaling projections for oxide TFTs that treat the band-tail DOS as a fixed material property will misestimate threshold voltage in short- and long-channel devices; the DOS must be entered as $D_{\mathrm{eff}}(E,L)$.
  • The same film can appear more disordered purely because it was patterned with a longer channel, so material comparisons of band-tail quality need to specify geometry.
  • At cryogenic temperatures, localization freezes out the disorder-dominated tail states first, meaning cooling amplifies threshold-voltage shifts and on-resistance in oxide TFTs.
  • Process levers that reduce localization—thicker films, O$_2$ annealing, and Ga/Zn substitution—give concrete paths to recover conventional linear scaling in scaled oxide and other low-dimensional channels.
  • Because eFPT maps the DOS directly in a working transistor, it can expose disorder-dominated transport in any thin-body FET, not just oxide semiconductors.

Reading between the lines

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

  • A direct test of the energy-alignment assumption would be to re-derive the $D_{\mathrm{eff}}$ versus $L$ plot after calibrating each device's chemical-potential axis against its own depletion-onset voltage; if the exponential decay were an alignment artifact, the corrected plot would flatten.
  • The same eFPT measurement could be applied to amorphous or polycrystalline 2D-material FETs to see whether the extracted $\xi(E)$ tracks independent measures of disorder such as the optical Urbach energy.
  • One implication the authors do not develop is that a length-dependent DOS makes the threshold voltage itself scale-dependent, so this localization-induced roll-off should be included in variability analysis of future back-end-of-line integrated oxide transistors.
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

4 major / 5 minor

Summary. The manuscript reports a lock-in-based electric-field penetration technique (eFPT) to map the effective density of states (DOS) in dual-gate InOx and IGZO thin-film transistors. The authors claim that the extracted effective DOS follows D_eff(E,L) = D0(E) exp[-L/xi(E)] (Eq. 5), which they interpret as size-dependent band-tail localization. They further argue that this localization explains a threshold-voltage roll-off mechanism, and they show temperature dependence and process/composition trends (thickness, O2 annealing, In concentration) that they interpret as systematic suppression of disorder. The central quantitative finding is the exponential channel-length dependence of the effective DOS at fixed chemical potential.

Significance. If the exponential length dependence of the effective DOS were established, it would be a notable contribution: it would show that transport-active band-tail states are not a material constant but depend on device geometry, with implications for scaling of amorphous oxide transistors. The eFPT technique itself, resolving four distinct electronic regimes (deep traps, shallow tail, disorder-dominated tail, extended states), is a useful experimental advance. The paper also provides a plausible engineering narrative (reducing localization by thickness, annealing, and Ga/Zn doping). However, the central quantitative claim rests on an untested chemical-potential alignment assumption and on fits without reported statistics; the supporting conductance-scaling cross-check is performed on the same devices and does not remove the risk of artifact. The idea is significant but the evidence as presented is not yet convincing.

major comments (4)
  1. [Geometric- and Temperature-Dependent Effective DOS, Eq. (5)] The chemical-potential alignment assumption is load-bearing and untested. The text states: 'The chemical potential for different channel sizes is referenced to zero gate bias, assuming a common intrinsic chemical-potential reference for devices fabricated from the same film.' If the flat-band or depletion-onset voltage varies with channel length (due to lateral film nonuniformity, doping gradients, or process-induced fixed charge), then the same nominal mu corresponds to different physical energies in different devices. For an intrinsic exponential Urbach tail, a linear-in-L offset in the energy reference produces an apparent D_eff(L) = D0 exp(-L/xi) with no localization at all. The manuscript does not provide an independent energy reference (e.g., depletion-onset edge, C-V flat-band, or thickness/doping calibration) for the different devices. Without such a check, Eq. (5) is not established; the authors should either validate the common-reference assumption experimentally or re-analyze the data at fixed back-gate voltage and demonstrate that the exponential suppression persists.
  2. [Figure 3d and Figure S3] The fits that support the exponential decay do not include error bars, confidence intervals, or goodness-of-fit statistics. The data in Fig. 3d appear to consist of five channel lengths (10–50 um) with no indication of device-to-device spread or measurement uncertainty. The extracted xi(E) values in Fig. S3 are therefore not accompanied by any uncertainty estimate. Given that the entire localization claim rests on the slopes of these lines, the authors should provide per-energy-point standard deviations, fit residuals, and the sensitivity of xi(E) to the choice of energy alignment and to the inclusion/exclusion of individual devices.
  3. [Figure S4 and cross-check of localization length] The 'independent' localization length extracted from conductance scaling in Fig. S4 does not provide a fully independent confirmation: it is obtained from the same devices, uses the same exponential fitting form, and shares the assumption that transport is dominated by a single localization length. The manuscript reports only that the two extractions agree 'within the same order of magnitude,' which is a weak quantitative check. A more convincing independent test would be a different measurement or analysis (e.g., magnetotransport, temperature-dependent resistance scaling with a known functional form, or a percolation-based model) that does not presuppose the exponential form of Eq. (5). As it stands, the confirmation is in part circular.
  4. [Equations (2)-(3), depletion-limit normalization] The extraction of n_s, mu, and D_eff relies on the full-depletion regime as a zero-density reference (V_dep). The manuscript does not state how V_dep is determined for each device and whether the depletion plateau is actually reached in all devices and at all temperatures. If V_dep varies with channel length (for instance, due to a length-dependent parasitic capacitance or incomplete depletion), the integration limits in Eq. (3) change, producing an artificial length dependence in the effective DOS and chemical potential. The authors should specify the criterion used to identify V_dep, show the raw I_TG/V_BG curves over the full gate-voltage range for every device, and quantify the robustness of the extracted D_eff to the choice of V_dep.
minor comments (5)
  1. [Figure 5c and text] The IGZO composition notation is inconsistent: the main text states In:Ga:Zn ratios of 5:1:1 and 7:1:1, while the Figure 5c caption refers to 'IGZO 311 TFTs.' Please clarify which ratio corresponds to '311' and unify the notation throughout.
  2. [Figure 3d] Please indicate whether the plotted points represent single devices or averages over multiple nominally identical devices; if multiple devices were measured, include error bars and the number of devices per channel length.
  3. [Throughout] The term 'disorders' is used as a countable noun (e.g., 'suppress disorders'), which is nonstandard in this context; consider using 'disorder' or 'disorder effects' for clarity.
  4. [Abstract and text] The phrase 'partially transport-active' is not precisely defined. Please specify the operational definition (e.g., the fraction of states contributing to conduction at a given energy, or the criterion used to separate transport-active from screening-active states).
  5. [Figure S4] The temperature and gate-voltage range of the conductivity scaling data are not stated in the main text or the SI caption. Since the localization length is energy-dependent and temperature-dependent, please include these measurement conditions.

Circularity Check

2 steps flagged · score 6.0 of 10

Central size-dependent DOS law is fixed by an assumed common mu reference; without it, any L-linear flat-band shift plus an Urbach tail yields the same exponential form.

  1. self definitional [Sec. 'Geometric- and Temperature-Dependent Effective DOS' (Eqs. 4–5, Fig. 3d)]
    "The chemical potential for different channel sizes is referenced to zero gate bias, assuming a common intrinsic chemical-potential reference for devices fabricated from the same film. ... The effective DOS decreases exponentially with increasing channel length, a hallmark of strong localization, and follows: D_eff(E,L) = D0(E) exp[-L/xi(E)] (5)."

    Eq. 4 fixes only dmu/dV_BG; the absolute mu origin is set by convention to V_BG=0 for every channel length. If the flat-band or depletion-onset voltage shifts with L (doping gradients, thickness nonuniformity, process charge), the same nominal E samples intrinsic energy E - delta(L). With an exponential Urbach tail D_int(E) ~ exp(E/E0), the constructed D_eff at fixed nominal E is D_int(E) exp[-delta(L)/E0]; any delta(L) = alpha L reproduces Eq. 5 exactly with 1/xi = alpha/E0. Thus Eq. 5 and the fitted xi are produced by the chosen energy-axis definition plus an intrinsic exponential tail, and are then labeled localization. The agreement with conductance scaling (Fig. S4) uses the same devices and same V_BG reference, so it does not independently discriminate.

  2. other [Abstract and Sec. 'Direct DOS Mapping by Electric-Field Penetration Technique' (Eq. 2)]
    "The extracted quantum capacitance, carrier density, and chemical potential reveal a disorder-dominated transport regime in which band-tail states are not merely passive traps, but become screening-active and partially transport-active."

    eFPT measures C_q = e^2 d n_s/d mu = e^2 D(E), i.e., it detects by definition only states that screen the penetration field. Calling the measured band-tail states 'screening-active' restates what the probe measures rather than an empirical discovery; states that do not screen are invisible to eFPT, so the contrast 'not merely passive traps' is not something the measurement can establish. This interpretive step is definitional, though it is secondary to Eq. 5.

full rationale

The main quantitative result D_eff(E,L)=D0(E) exp[-L/xi(E)] is not derived from an independently calibrated energy scale. The paper's own Eq. 4 yields mu only up to a constant, and the constant is fixed by the stated assumption that all devices share mu=0 at V_BG=0. Since the intrinsic oxide band tail is exponential (Urbach), a channel-length-dependent flat-band offset—unexcluded by the data—turns that convention into exactly the exponential-in-L law reported, making xi a fit parameter of the alignment convention. The conductance-scaling check on the same devices, using the same V_BG reference, is not an external benchmark. The temperature- and process-trend data are coherent with disorder reduction and would survive a corrected energy reference, so this is partial rather than total circularity. The 'screening-active' claim is tautological with the eFPT definition of C_q. Reference [8] (same group) is cited for the exponential-resistance premise, but the main evidence does not reduce to that citation; accordingly this is not scored above 6.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central measurement is the quantum capacitance derived DOS, so the main levers are the capacitor model, the depletion reference, and the assumed common chemical-potential alignment. The localization interpretation leans on a self-cited prior framework and on exponential fits with fitted xi and D0. No new physical entities are proposed.

free parameters (3)
  • xi(E) (localization length) = not quoted; plotted in Fig. S3
    Extracted from the slope of linear fits of ln D_eff vs channel length in Fig. 3d/Eq. 5. It carries the interpretation of the localization origin and is compared only loosely with conductance-derived xi.
  • D0(E) (short-channel-limit DOS) = not quoted; intercept of fits
    The intercept of the exponential fit in Eq. 5; used to normalize the effective DOS but not independently verified.
  • Geometric scaling parameters in Fig. 5d = not quoted
    Slopes summarizing the exponential suppression for different films and process conditions; fitted values underpin the disorder-engineering claims.
assumptions (5)
  • domain assumption Dual-gate capacitor model with I_TG/I_TG0 = (C_TG+C_BG)/(C_TG+C_BG+C_q)
    The extraction assumes the top-gate current is pure displacement current through the channel and that the channel is thin enough for field penetration; stated in the equivalent circuit of Fig. 2b and Eq. 2.
  • standard math Quantum capacitance relation C_q = e^2 d n_s/d mu = e^2 D(E)
    Standard compressibility relation used in Eq. 2; assumed valid in the presence of disorder.
  • domain assumption Full depletion regime is the zero-carrier reference (I_TG^0)
    The method normalizes by the depletion-limit current and integrates from this reference (Eq. 3); if the channel is not fully depleted, n_s and mu are offset.
  • ad hoc to paper Common intrinsic chemical-potential reference across devices of different lengths
    Explicitly stated in the section on geometric dependence; needed to compare D_eff at fixed mu across devices. Untested and load-bearing.
  • domain assumption Disorder-dominated regime with L > xi gives exponential resistance growth and localized band-edge states
    Adopted from the authors' prior work (ref 8, Niu et al., Nano Lett. 2026). The paper treats it as given and uses it to interpret the data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Size-Dependent Band-Tail Localization in Oxide Semiconductors Revealed by Direct Density-of-States Mapping." pith.science (2026). https://pith.science/paper/OAHO7PR7

@misc{pith2026260809283,
  author       = {Pith},
  title        = {Pith review of: Size-Dependent Band-Tail Localization in Oxide Semiconductors Revealed by Direct Density-of-States Mapping},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OAHO7PR7}},
  note         = {Machine review of arXiv:2608.09283}
}
read the original abstract

Disorder-induced localization is expected to become increasingly important as amorphous oxide semiconductor transistors are scaled toward low-dimensional channels, yet the electronic states responsible for this transport regime remain difficult to resolve experimentally. Here, we use a lock-in-based electric-field penetration technique to directly map the effective density of states (DOS) in In-based oxide semiconductor thin-film transistors (TFTs). The extracted quantum capacitance, carrier density, and chemical potential reveal a disorder-dominated transport regime in which band-tail states are not merely passive traps, but become screening-active and partially transport-active. Geometry-dependent DOS mapping shows an exponential suppression of the effective DOS with channel length, demonstrating size-dependent band-tail localization and providing a microscopic origin for a distinct localization-induced threshold-voltage roll-off mechanism. Temperature-dependent measurements show that the disorder-dominated DOS is strongly suppressed at low temperatures, while extended diffusive states remain nearly unchanged, confirming the localization origin. By tuning film thickness, O2 annealing, and In/Ga/Zn composition, we further demonstrate systematic suppression of disorder and effective-DOS localization. This work establishes direct DOS mapping as a device-level probe of localization physics and provides a pathway for engineering disorder in low-dimensional oxide semiconductor electronics.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

4 extracted references · 1 canonical work pages

  1. [1]

    and McConville, C.F., Surface electron accumulation and the charge neutrality level in In2O3

    (1) King, P.D.C.; Veal, T.D.; Payne, D.J.; Bourlange, A.; Egdell, R.G. and McConville, C.F., Surface electron accumulation and the charge neutrality level in In2O3. Physical review letters 2008, 101(11), p.116808. https://doi.org/10.1103/PhysRevLett.101.116808. (2) Belmonte, A. ; Oh, H. ; Rassoul, N. ; Donadio, G.L. ; Mitard, J. ; Dekkers, H. ; Delhougne,...

    arXiv 2008

  2. [170]

    (4) Si, M.; Hu, Y.; Lin, Z.; Sun, X.; Charnas, A.; Zheng, D.; Lyu, X.; Wang, H.; Cho, K.; Ye, P

    https://doi.org/10.1038/s41928-022-00718-w. (4) Si, M.; Hu, Y.; Lin, Z.; Sun, X.; Charnas, A.; Zheng, D.; Lyu, X.; Wang, H.; Cho, K.; Ye, P. D. Why In 2O3 Can Make 0.7 nm Atomic Layer Thin Transistors. Nano Lett. 2021, 21, 500–506. https://doi.org/10.1021/acs.nanolett.0c03967. (5) Niu, C.; Tan, P.; Lin, J.-Y.; Long, L.; Lin, Z.; Zhang, Y.; Wang, H.; Wilk,...

    arXiv 2021

  3. [2012]

    Conduction in non -crystalline materials: III

    (13) Mott, N.F. Conduction in non -crystalline materials: III. Localized states in a pseudogap and near extremities of conduction and valence bands. Philosophical Magazine, 1969, 19(160), pp.835-852. https://doi.org/10.1080/14786436908216338. (14) Anderson, P.W. Absence of diffusion in certain random lattices. Physical review, 1958, 109(5), pp.1492-1505. ...

    work page arXiv 1969

  4. [2025]

    (27) Park, H.; Gopal Kirtania, S.; Sarkar, E.; Chakraborty, D.; Zhang, C.; Jae Lee, H.; Shin, J.; Yu, S.; Khan, A.; Kim, H.; Im, C.; Hong, M

    https://doi.org/10.23919/VLSITECHNOLOGYANDCIR65189.2025.11075003. (27) Park, H.; Gopal Kirtania, S.; Sarkar, E.; Chakraborty, D.; Zhang, C.; Jae Lee, H.; Shin, J.; Yu, S.; Khan, A.; Kim, H.; Im, C.; Hong, M. J.; Ha, D.; Datta, S. Extreme EOT Scaling in Tungsten-Doped In2O3 MOSFETs for Enhanced Stability and Drive Current. IEEE Trans. Electron Devices 2025...

    arXiv 2025

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.