{"id":"03c89b50-6dee-4bbc-86b1-f8551c5107d0","arxiv_id":"2507.14213","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Voltage-controlled nitrogen migration in FeCoN dots creates probabilistic magnetic states that the authors use to demonstrate true random number generation and physical unclonable functions.","lead":"This paper builds a prototype hardware security device from tiny magnetic dots whose magnetic state can be switched by voltage, and shows the states can be random or tunable. The idea is to use material randomness itself as a source of secret keys and device fingerprints.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"TRNG/PUF claims rest on an untested assumption that degaussed states are independent and stationary; the reported FHDintra≈0.5 is computed from a post hoc binary grouping and cannot substitute for a formal randomness test.","rationale":"The reader's weakest assumption—that AC degaussing yields independent, identically distributed state choices with stable probabilities—is indeed the load-bearing point. My concern sharpens this: the only evidence offered for independence, FHDintra≈0.5, is computed after a post hoc binary grouping of the four physical states, and the raw four-state FHDintra (0.505) actually deviates from the expected 0.75. This means the apparent binary randomness is not an independent confirmation but partly a consequence of the chosen grouping. Furthermore, FHDintra and average Shannon entropy do not test temporal stationarity or serial independence, which are necessary for TRNG and PUF security. The concrete test would settle whether the assumption holds. This concern does not negate the experimental proof-of-concept of selective actuation and probabilistic magnetic states; it does place a stronger condition on the security-primitive claims. Since the reader already conditioned acceptance on similar missing evidence, the recommended verdict remains CONDITIONAL, i.e., UNCHANGED relative to the reader's assessment.","tokens_in":19701,"tokens_out":2982,"duration_ms":37362,"concrete_test":"Take the raw per-cycle bit assignments for the 24-bit circuit (and the 18-bit circuit) from the 100 degauss/imaging cycles, in the order collected. Compute: (1) min-entropy per bit rather than average Shannon entropy; (2) lag-1 through lag-10 autocorrelation for each bit and a chi-square test against the fixed Bernoulli model with the enrolled probability; (3) pairwise mutual information between dots to detect cross-dot correlations; (4) stability by splitting the 100 cycles into two halves and testing whether per-bit probabilities are equal within sampling error. If min-entropy per bit falls below, say, 0.98 or any tested correlation is significant at the 0.05 level after multiple-comparison correction, the claimed 'absence of correlations' and near-perfect randomness are not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central security claims require that, after AC degaussing, each dot's orientation/chirality bit is independent across dots and across repeated cycles, with probabilities that are stable over time. The evidence offered is Eq. (4) FHDintra and Eq. (6) average Shannon entropy, both computed from 100 enrolment cycles. This is insufficient in two ways. First, the binary grouping used to obtain FHDintra≈0.5 (grouping states 1+4 versus 2+3) is introduced after the data are seen; the four-state FHDintra is 0.505, far from the ideal 0.75, showing that SD/vortex probabilities are strongly imbalanced. A grouping chosen to maximize apparent randomness cannot independently confirm 'absence of correlations between degaussed states' (Fig. 3k). Second, neither FHDintra nor average Shannon entropy tests for serial autocorrelation, cross-dot dependence, or stationarity of probabilities over time. Entropy is reported as average Shannon entropy per bit, not min-entropy; genuine unpredictability for cryptographic use requires min-entropy and validated independence. Without a formal randomness assessment (e.g., NIST SP 800-90B), a lag-k autocorrelation check, and a stationarity test across cycle subsets, the TRNG and PUF claims are not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a magneto-ionic approach to hardware security primitives based on selectively voltage-actuated FeCoN dots. The authors show that negative voltage gating drives N3- ion migration in pre-patterned dots, generating ferromagnetic FeCo sublayers whose thickness determines whether a dot behaves as a deterministic (single-domain or vortex) or probabilistic bit after AC degaussing. They report stochastic orientation/chirality of the degaussed states, voltage-tunable SD/vortex probabilities, and they propose applications as true random number generators, physical unclonable functions, and in-memory probabilistic inference. The experimental backbone includes selective actuation in two circuits, EELS cross-sections of contacted and non-contacted dots, Kerr imaging of the four magnetic states, and 100-cycle enrolment statistics for two samples.","tokens_in":19888,"tokens_out":6078,"duration_ms":72940,"significance":"The experimental demonstration of selective magneto-ionic actuation in patterned dots, with EELS evidence of the planar nitrogen migration front and Kerr imaging of SD/vortex coexistence, is a genuine and useful contribution to magneto-ionics. The observation that voltage actuation time alters the SD/vortex probability balance is a falsifiable materials result with potential for security and stochastic-computing applications. The paper is also transparent in providing the enrolled probability tables and detailed methods. However, the security claims currently outrun the statistical evidence: the randomness metrics are computed in-sample from a post hoc binary grouping, the entropy reported is average Shannon entropy rather than min-entropy, no formal randomness test or stability study is provided, and PUF uniqueness is assessed on only two samples with post hoc CRP selection. The underlying materials science is sound, but the TRNG/PUF/inference claims require substantial additional validation to meet cryptographic standards.","major_comments":[{"comment":"The claim that degaussed states are 'truly random and uncorrelated' rests on a post hoc binary grouping of the four observed states (grouping states 1+4 and 2+3). The four-state FHDintra is 0.505, far from the ideal 0.75 for four equiprobable states, showing strong marginal imbalance. A binary FHDintra close to 0.5 only indicates that the two constructed subclasses are roughly balanced; it does not test independence across dots, serial autocorrelation across degauss cycles, or stationarity of the probabilities over time. No formal randomness test (e.g., NIST SP 800-90B), no lag-k autocorrelation analysis, and no long-term stability check are reported. Because the TRNG and PUF claims depend on independent and stable bits, this is a load-bearing gap.","section":"Section 'Unveiling probabilistic and stochastic behaviour' and Fig. 3k"},{"comment":"The entropy metric used is the average Shannon entropy per bit, not min-entropy, which is the relevant quantity for cryptographic unpredictability. For example, Supplementary Table 1 lists dot 1 with p_R,CW = 61.29%, giving a per-bit min-entropy of about 0.71 bits rather than the reported Shannon entropy of 0.963. Additionally, the probabilities p_i are estimated from the same 100 enrolment cycles used to compute FHDintra, so the reported H_i = 0.97 is an in-sample estimate without held-out validation. The relationship S = 2^{H_total} = f(N) also conflates the support size of the output distribution (2^N) with the effective number of typical sequences (2^{N H}); the password-space counts for N=24, 42, and 100 dots are therefore not justified by the data as presented.","section":"True random number generator section and Eqs. (5)-(6)"},{"comment":"The CRP counts and BER values are computed from the enrolled probabilities rather than measured on independent verification data. Table 1 reports BER for T = 1, 5, 11, and 27 as theoretical binomial predictions, but no empirical error rates on fresh degauss/imaging cycles are provided. Equation (7) counts 2^m distinct responses for m p-bits, which treats every combination of possible outcomes as a valid response; because a p-bit does not reliably produce a specific response, this inflates the CRP count. An empirical BER evaluation on a held-out set and a clear response-binarization rule are needed to support the PUF and inference claims.","section":"Probabilistic device fingerprints, Eqs. (7)-(10) and Table 1"},{"comment":"The inference demonstration uses the same probability library to select the five-dot CRP and to evaluate classification, and the manuscript does not state whether the 27 evaluation trials are disjoint from the 100-cycle enrolment dataset. If the trials overlap, the reported ~90% inference reliability in Fig. 4k is an in-sample result. Moreover, no false-acceptance or false-rejection rates are reported, and a 90% success rate after 27 iterations is not obviously sufficient for authentication without a threshold analysis. The protocol should be evaluated on fresh degauss/imaging cycles acquired after enrolment.","section":"Probabilistic inference protocol and Supplementary Table 4"},{"comment":"PUF uniqueness is not experimentally established. Only two samples are compared, and the pFHDinter values are computed from the two samples' enrolled probabilities rather than measured on a population of devices. The 'strong uniqueness' combinations (pFHDinter ≈ 0.5) are selected after examining the probability libraries, which introduces selection bias. To support the PUF claim, the authors should measure inter-device Hamming distances across multiple independently fabricated devices using fixed, pre-defined challenges, and report the distribution of those distances without post hoc selection.","section":"PUF uniqueness assessment, pFHDinter and Table 1"}],"minor_comments":[{"comment":"The notation H_total = -Σ H_i is confusing because H_i is already a nonnegative per-bit entropy; the total should simply be H_total = Σ H_i.","section":"Eq. (5)"},{"comment":"The phrase 'truly random and uncorrelated' overstates the evidence; a more measured wording such as 'consistent with uncorrelated binary subclasses' would be appropriate given the post hoc grouping and the absence of formal randomness tests.","section":"Fig. 3k caption"},{"comment":"The reported probabilities are point estimates from 100 cycles; with N=100, the standard error of a 50% probability is 5%. Adding binomial confidence intervals or error bars would clarify how much the entropy, BER, and pFHDinter values could vary.","section":"Supplementary Tables 1-3"},{"comment":"The statement about 'quantum randomness generation' is not supported by any experiment in the paper; either remove this claim or explicitly label it as a future prospect.","section":"Conclusions"},{"comment":"The term 'fully selective' voltage actuation is stronger than what is demonstrated: only two circuits per array are selectively activated. A softer wording, e.g., 'selective', would better match the experimental evidence.","section":"Abstract and Introduction"},{"comment":"The binomial coefficients in Eq. (7) use symbols C and H without definition in the main text; the roles of C (deterministic bits?) and H (p-bits?) should be defined explicitly before the equation.","section":"Eq. (7)"}],"recommendation":"major_revision","confidential_remarks":"The manuscripts' core materials-science finding - selective voltage actuation of FeCoN dots with voltage-tuneable SD/vortex probabilities - is solid and original. The security framing, however, is currently not validated to the level claimed: the randomness, keyspace, BER, and uniqueness metrics are in-sample or rely on post hoc selection. I believe these gaps are fixable within a revision by adding formal randomness tests (e.g., NIST SP 800-90B on the bitstreams), computing min-entropy, performing held-out verification cycles, and measuring inter-device distances on multiple samples. If the authors prefer not to add those experiments, the security claims should be correspondingly softened. I would not reject the paper, as the material demonstration is valuable for the magneto-ionics community, but the security-oriented claims need to be brought in line with the evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read this one for the materials physics, not for the security guarantees. The genuinely new result is selective voltage-driven N3− migration in patterned FeCoN dots: individual circuits can be actuated while neighboring dots stay paramagnetic, and the resulting dots show voltage-tunable probabilities between single-domain and vortex states after degaussing. The EELS cross-sections and Kerr imaging support the actuation claim, and the observation of stable d-bits versus p-bits at fixed dot positions is a real proof-of-concept. The authors also deserve credit for reporting the raw numbers that undermine their own headline: the four-state FHDintra is 0.505, not the 0.75 expected for independent four-state draws, and they openly acknowledge 40% initial inference accuracy on the hardest CRP before majority voting.\n\nThe soft spots are all on the security side, and they are substantial. The FHDintra≈0.5 used to claim 'absence of correlations' comes from a post hoc binary grouping (1+4 vs 2+3) chosen after seeing the data; the same enrolled probabilities are used both to define the grouping and to compute entropy, so the near-0.5 result is partly baked in. The entropy figure is average Shannon entropy per bit, not min-entropy, and no NIST SP 800-90B test, autocorrelation check, or stationarity analysis across cycle subsets is provided. The TRNG and PUF claims therefore rest on an untested assumption that degaussed states are independent across dots and over time. That may be true, but the paper doesn't show it. The 100-bit password estimate scales from average entropy and ignores drift and correlations. The probabilistic inference demonstration is also a cherry-picked CRP: they select five-dot combinations with pFHDinter near 0.5, which is fine for illustration, but the 90% figure applies to one selected combination after 27 averaging iterations, not to general authentication.\n\nNone of this kills the core contribution. The selective actuation and voltage-tunable probability data are new and credible, and the limitations are addressable with more measurements and a more careful statistical treatment. But the paper should not be published as a TRNG/PUF security result in its current form. It needs a revision that reports min-entropy and formal randomness tests, a stationarity check (e.g., first vs second half of enrolment), the ungrouped four-state analysis without post hoc grouping, and error bars on all entropy/FHD values. If those are added, this could be a useful security-hardware paper rather than a materials paper with security adjectives.\n\nWho is this for? People working on magneto-ionics and emerging hardware security primitives, especially those interested in reconfigurable physical randomness. It deserves peer review, but a serious referee should make the authors do the statistical work.","headline":"A credible magneto-ionic proof-of-concept whose security claims need more statistical rigor before it can be taken as a TRNG/PUF.","tokens_in":20518,"tokens_out":2979,"would_cite":true,"duration_ms":33176,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Voltage-driven nitrogen migration in FeCoN dots turns the material itself into true random number generators, physical unclonable functions, and probabilistic inference devices.","keywords":["magneto-ionics","hardware security","true random number generator","physical unclonable function","probabilistic computing","FeCoN dots","nitrogen ion migration","magnetic vortex states"],"falsifier":"Record a long stream of consecutive degaussing outputs from one array and apply a standard randomness test suite plus an autocorrelation check; a significant lag-one correlation, a failed test, or a per-dot vortex probability that drifts outside its enrolment confidence interval would falsify the true-randomness and PUF claims.","tokens_in":19417,"feed_emoji":"🧲","tokens_out":9266,"duration_ms":101965,"temperature":0.7,"pith_summary":"This paper tries to establish that a patterned magnetic material can serve as its own security hardware: by applying a negative voltage to selected FeCoN (iron–cobalt–nitride) dots, nitrogen ions migrate out in a planar front, converting normally paramagnetic dots into ferromagnetic dots whose magnetic sublayer thickness is set by the gating time. After an alternating-field demagnetisation (degaussing), each dot settles into one of two single-domain orientations or one of two vortex chiralities, with some dots making the same deterministic choice every time and others switching probabilistically between single-domain and vortex states. The paper argues that the random orientation and chirality draws are uncorrelated (a state-difference metric of 0.474 against an ideal 0.5) and nearly maximal in information content (about 0.97–0.99 bits of entropy per bit position), while the voltage-tuneable fraction of single-domain versus vortex states provides a controllable probabilistic fingerprint. Taken together, these behaviours are used to demonstrate true random number generation, physical unclonable function authentication, and probabilistic inference in a proof-of-principle dot array. A sympathetic reader would care because this would embed randomness, tamper resistance, and reconfigurability directly into the material rather than in an attached processor.","feed_headline":"Ion motion turns magnetic dots into unclonable security primitives","feed_subtitle":"Random bits and tunable fingerprints for cryptography and anti-counterfeiting live inside the material.","key_machinery":"The load-bearing mechanism is voltage-controlled nitrogen migration in FeCoN dots, described by the paper as a planar N3− ion migration front. Negative gate voltage expels nitrogen from the bottom of each dot, forming a ferromagnetic FeCo sublayer of tuneable thickness; that thickness selects between single-domain and vortex ground states, while the orientation or chirality of the resulting state is set stochastically by an alternating-field demagnetisation step. Selective gold contacts allow the voltage to be applied to chosen dots only, so part of an array can be switched on while the rest stays paramagnetic. The readout is magneto-optic imaging of the two in-plane magnetisation components, which assigns each dot one of four states; repeated degaussing and imaging cycles generate the probability libraries, state-difference statistics, entropy estimates, and bit-error rates that the security claims are built on.","core_discovery":"On the paper's own terms, the discovery is that selective voltage actuation of FeCoN dots—not just blanket gating of continuous films—produces a reconfigurable set of magnetic states rich enough to serve as hardware security primitives. Negative gate voltage drives N3− ions out of the dot in a planar front, leaving a ferromagnetic FeCo sublayer whose thickness grows with gating time; for a 20-nm dot this thickness decides whether the dot's ground state after degaussing is a single domain or a vortex. Some dots are deterministic, always returning to the same state, while others are probabilistic, with a voltage-tuneable probability of single-domain versus vortex; in both classes the in-plane orientation or chirality is random from one degaussing cycle to the next. The paper demonstrates that a 24-dot array gives roughly 0.97 bits of information-theoretic entropy per bit and about 9.84×$10^{6}$ possible random sequences, that enrolling such probabilities creates device fingerprints with tens of thousands of challenge–response pairs, and that a five-probabilistic-bit challenge can authenticate a sample with about 90% confidence after 22–27 repeated reads. The central proposition is that the material itself, rather than any software or attached circuitry, carries the secret.","pith_inferences":["The paper's mechanism suggests that replacing the optical readout with an on-chip magnetoresistive or Hall sensor would let the same dot array act as a compact PUF and TRNG without a microscope; this is an integration step the paper does not attempt.","The voltage-tuneable single-domain versus vortex probability could be viewed as a physical probabilistic bit whose bias is set electrically, so the same array might be repurposed for Bayesian or stochastic computing alongside security.","A direct test would be to run a standard randomness test suite on long streams of successive degaussing outputs; because individual trials take about 3.5 seconds, a few hours of data would reveal any hidden serial correlation or drift that the 100-cycle enrolment cannot see.","If the planar-front assumption transfers to smaller dots, the same security concept could be scaled to much denser arrays; measuring whether the migration front remains planar below roughly one micrometre would decide that question."],"forward_implications":["If the probabilities stay stable, a 100-dot magneto-ionic lock would admit about 3.17×10^29 passwords at 0.98 bits per bit, making billion-guesses-per-second brute force impractical for more than the age of the universe.","Shortening the gating time from 60 to 30 minutes raises the average single-domain probability from about 10.7% to 28.2%, turning more dots into probabilistic bits and increasing the available challenge–response pairs from about 41,000 to 60,000; voltage duration is therefore a reconfigurability knob for the fingerprint.","Because the probabilities live in the material, stray or applied magnetic fields are erased by the next degaussing cycle, so the device can only be reconfigured electrically; attempted voltage tampering switches previously off dots, exposing the intrusion.","A five-probabilistic-bit readout can distinguish one sample from another by majority voting over repeated degaussing cycles, reaching roughly 90% certainty in about 1.5 minutes without any electrical connection during readout."],"supporting_citations":[{"why":"Establishes the voltage-driven single-domain to vortex transition in FeCoN and the planar nitrogen migration front that the patterned dots extend to selective actuation.","marker":"20"},{"why":"Shows voltage-driven motion of nitrogen ions as a general magneto-ionic paradigm that grounds the claim that N3− migration is the active mechanism.","marker":"26"},{"why":"Reports the ternary nitride FeCoN system whose magneto-ionic response and optimisation the present dots inherit.","marker":"32"},{"why":"Provides the size- and thickness-dependent single-domain versus vortex behaviour used to interpret the observed dot states.","marker":"33"},{"why":"Argues that random phenomena in magnetic materials can serve data security and computing, forming the conceptual basis for using stochastic magnetic states as security primitives.","marker":"19"},{"why":"Links stochastic memory devices to TRNG and security applications, providing the framework in which the paper's entropy and bit-error metrics are meaningful.","marker":"8"}],"fun_headline_variants":["Voltage moves ions to make unclonable magnetic keys","Ion-driven magnetic states enable tamper-proof cryptography","Voltage-gated ion migration yields security primitives in dots","FeCoN dots with voltage-tuned magnetism for hardware security","Ion motion creates random magnetic bits for anti-counterfeiting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole construction rests on the assumption that repeated AC demagnetisation gives each dot an independent random choice of orientation or chirality with a probability that stays stable over time, and that the planar nitrogen-migration behaviour established for continuous films also holds in the patterned dots.","fun_headline_variants_meta":{"raw":{"variants":["Voltage moves ions to make unclonable magnetic keys","Ion-driven magnetic states enable tamper-proof cryptography","Voltage-gated ion migration yields security primitives in dots","FeCoN dots with voltage-tuned magnetism for hardware security","Ion motion creates random magnetic bits for anti-counterfeiting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000636,"raw_usage":{"total_tokens":2960,"prompt_tokens":1002,"completion_tokens":1958,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":618,"completion_tokens_details":{"reasoning_tokens":1874}},"tokens_in":618,"tokens_out":1958,"duration_ms":18157,"temperature":1.0,"reasoning_tokens":1874,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:08:49.124489+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record a long stream of consecutive degaussing outputs from one array and apply a standard randomness test suite plus an autocorrelation check; a significant lag-one correlation, a failed test, or a per-dot vortex probability that drifts outside its enrolment confidence interval would falsify the true-randomness and PUF claims.","supporting_citations":[{"cited_title":"K., Sort, J","cited_arxiv_id":null,"evidence_quote":"Establishes the voltage-driven single-domain to vortex transition in FeCoN and the planar nitrogen migration front that the patterned dots extend to selective actuation."},{"cited_title":"O., Butterling, M., Wagner, A., Sireus, V ., Abad, L., Jensen, C","cited_arxiv_id":null,"evidence_quote":"Shows voltage-driven motion of nitrogen ions as a general magneto-ionic paradigm that grounds the claim that N3− migration is the active mechanism."},{"cited_title":"O., Butterling, M., Attallah, A","cited_arxiv_id":null,"evidence_quote":"Reports the ternary nitride FeCoN system whose magneto-ionic response and optimisation the present dots inherit."},{"cited_title":"P., Koltsov, D","cited_arxiv_id":null,"evidence_quote":"Provides the size- and thickness-dependent single-domain versus vortex behaviour used to interpret the observed dot states."},{"cited_title":"Exploiting random phenomena in magnetic materials for data security, logics, and neuromorphic computing: Challenges and prospects","cited_arxiv_id":null,"evidence_quote":"Argues that random phenomena in magnetic materials can serve data security and computing, forming the conceptual basis for using stochastic magnetic states as security primitives."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Links stochastic memory devices to TRNG and security applications, providing the framework in which the paper's entropy and bit-error metrics are meaningful."}],"review_version":1}