{"id":"eadcd809-e466-4673-9ad8-35ff4930bd4b","arxiv_id":"2506.01409","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Iterative differential deposition of WSi2 films corrected 300 mm silicon x-ray mirrors from several nm to as low as 0.1 nm RMS height error over 260 mm.","lead":"Differential deposition of WSi2 films reduced surface height errors on four 300 mm silicon x-ray mirrors by factors of roughly 16 to 36, reaching residuals below 0.5 nm RMS. The work shows that two sputter coaters with different aperture sizes can iteratively flatten synchrotron mirror surfaces to the 0.1 nm range.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Untested static-to-dynamic kernel equivalence underlies the deconvolution; direct open-loop verification is needed before the 0.1 nm RMS residuals are treated as deterministic.","rationale":"I aligned with the reader's weakest assumption because Eq. (1) is the quantitative foundation of the method; all correction profiles are computed from it. The final residuals after a series of closed-loop iterations cannot by themselves validate the kernel, since a biased kernel may still converge to a low residual if the initial error is large and the bias is moderate. A single open-loop dynamic deposition test is the cleanest way to separate the forward model from the feedback loop. The paper otherwise presents credible, internally consistent experimental data: Table 3 confirms roughly 20-30x RMS improvement for the three WP mirrors (27.6, 35.6, 33.2) and below-0.5 nm RMS for all four mirrors; WP#37 reaches 0.11 nm RMS / 0.76 nm PV over 280 mm (0.095 / 0.47 over 260 mm). The metrology repeatability (<0.05 nm RMS) and the stated LTP cross-check give adequate support for sub-nanometer residuals, though an independent re-measurement of the final mirrors would further strengthen the paper. The acknowledged central-line-only nature of the correction and the second-order polynomial subtraction are disclosed in the text and do not change my conclusion. The reader's CONDITIONAL verdict should stand: the content is acceptable with the condition that the static-to-dynamic kernel assumption be validated (or the claims re-scoped) and that raw data/code be made available.","tokens_in":7726,"tokens_out":10180,"duration_ms":119945,"concrete_test":"On a fresh witness mirror, apply a single open-loop CMCS correction pass (1 mm aperture) using a velocity profile computed from Eq. (1) and the static kernel in Fig. 1(d), targeting a known 1 nm-amplitude, 10 mm-period test profile. Measure the deposited WSi2 thickness over the central 260 mm with Fizeau stitching and compare with the profile predicted by convolving the static kernel with the speed profile. If the residual mismatch exceeds 0.2 nm peak-to-valley or 0.05 nm RMS, the static-to-dynamic kernel assumption is falsified; agreement at that level would validate the deconvolution at the claimed accuracy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2.1 defines the method by Eq. (1): the dynamic thickness deposited on a moving substrate is predicted from the normalized static flux profile f(xm−xs), measured on stationary substrates in §3.1, and a constant central growth rate R. This equivalence is never tested. In a DC magnetron sputter system the effective emission pattern seen by a moving substrate can differ from the stationary aperture profile: the substrate samples different source angles, the plasma may drift within a duty cycle, and the 1–3 mm aperture-to-substrate spacing makes motion-dependent shadowing and position jitter (0.1–0.3 mm) non-negligible. A kernel mismatch enters the deconvolution as a systematic bias, not as noise; the same biased speed-profile calculation is then repeated on each iteration. The observed monotone convergence is reassuring and rules out a gross failure, but it does not settle the concern at the 0.1 nm level, because a biased kernel can still reduce a dominant initial figure error while leaving a consistent, kernel-dependent residual. Since the central claim is specifically about reaching 0.1 nm RMS and 0.5 nm PV down to 2 mm periods, the calculation must be correct at exactly that amplitude scale.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes an experimental development of differential deposition (DD) for correcting the figure of 300 mm flat silicon mirrors intended for x-ray beamlines. A deconvolution algorithm, built on Eq. (1), converts a measured height-error profile into a substrate velocity profile using the static deposition-flux kernel measured in Section 3.1. Corrections are applied iteratively on two DC magnetron sputter systems with apertures from 24 mm down to 1 mm, and the resulting surfaces are characterized by Fizeau stitching metrology. The authors report residual errors down to 0.1 nm RMS and 0.5 nm PV over a 260 mm clear aperture, PSD reductions in frequency bands corresponding to the aperture widths, and no significant roughness degradation. They conclude that DD is a practical sub-nanometer finishing method for long x-ray mirrors.","tokens_in":7962,"tokens_out":7931,"duration_ms":81735,"significance":"If the headline residual levels are taken at face value, the paper is a significant experimental contribution: it demonstrates iterative convergence of DD into the sub-nanometer regime on 300 mm mirrors, with cross-checks (LTP comparison, PSD analysis, roughness tracking) that strengthen the qualitative conclusions. The per-iteration data in Table 3 and Figure 5 give a transparent record of convergence and spread, and Figure 6 provides a clear link between aperture choice and corrected spatial-frequency band. The central technical caveat is that the deposition kernel is measured statically but applied dynamically, so the absolute accuracy of the 0.1 nm RMS claim rests on an assumption that is not directly tested.","major_comments":[{"comment":"The deconvolution in Eq. (1) uses the normalized static flux profile f determined from stationary-substrate depositions (Section 3.1, Fig. 1) as the deposition kernel during dynamic substrate motion. The manuscript does not provide a direct experimental test that this kernel is identical under dynamic conditions; plasma drift, motion-dependent shadowing, and the stated sample position accuracy of ±0.1–0.3 mm with aperture-substrate distances of 1–3 mm can all change the effective flux profile seen by a moving substrate. A kernel mismatch enters the computed velocity profile as a systematic bias, not as noise, and the same biased profile is reused in subsequent iterations, so the observed monotone convergence does not by itself validate the calculation at the 0.1 nm RMS scale claimed in Section 4. I recommend adding a validation experiment in which a known dynamic deposition profile is predicted from Eq. (1) and compared with the measured profile, or an equivalent test of the static-to-dynamic kernel equivalence.","section":"Sections 2.1 and 3.1"},{"comment":"The abstract states that shape errors were 'routinely reduced by a factor of 20-30', and Section 4 says 'about 30x (RMS) and 20x (PV)'. Table 3 gives improvement factors over 280 mm of 16.8–35.6x for RMS and 10.2–16.0x for PV; over 260 mm the PV factors are 8.1–24.7x. These tabulated values do not support a uniform '20-30x' claim and never support '20x PV'. The wording should be revised to report the measured ranges separately for RMS and PV and to note that one mirror (CO#7) was affected by contamination, which weakens the 'routine' characterization.","section":"Abstract, Section 4, Table 3"}],"minor_comments":[{"comment":"Equation (1) is rendered with a formatting artifact that makes the division by v(x_m) illegible; please provide a cleanly typeset equation and define the integration variable explicitly.","section":"Eq. (1)"},{"comment":"The improvement factors in Table 3 are computed with respect to iteration #1, which is the state after the initial uniform WSi2 coating, not the bare substrate. Although Section 3.2 explains that iteration #0 was omitted because the uniform coating did not change the figure, the table caption should state this baseline explicitly.","section":"Table 3 and Section 3.2"},{"comment":"The Fizeau repeatability is quoted as better than 0.05 nm RMS, which is non-negligible compared with the best residual values in Table 3 (e.g., 0.095 nm RMS for WP#37 over 260 mm). Please include an uncertainty estimate or at least state whether the quoted residuals are repeatability-limited.","section":"Section 2.4 and Table 3"},{"comment":"The roughness discussion would benefit from a quantitative statement of the measurement uncertainty for the Wyko RMS values, since the claimed 'virtually constant' behavior and the WP#32 increase rest on differences of order 0.1 nm.","section":"Section 3.4 and Fig. 7"},{"comment":"Residual profiles in Fig. 4 are shown on a common scale, but the spatial-frequency content that dominates each residual is not quantified; a short discussion or a mini-PSD panel would help support the statement that long-period undulations have weaker wavefront impact.","section":"Section 3.2 and Fig. 4"}],"recommendation":"major_revision","confidential_remarks":"The static-to-dynamic kernel validation is the key technical uncertainty; if the authors can provide a direct dynamic-kernel test or soften the headline claims to match Table 3, I would support publication. The paper is within scope for physics.optics and the experimental data are valuable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a solid engineering result. The ESRF group has pushed differential deposition to 0.1–0.2 nm RMS figure errors on 300 mm flat Si mirrors using the new compact coating system and a 1 mm aperture. The measured residuals are real, the PSD analysis shows frequency bands dropping exactly where the aperture widths predict, and the four-mirror series gives a useful sense of repeatability. The genuinely new content is the CMCS coater, the narrower aperture, and the demonstrated sub-0.2 nm level—previous work from this group was at the sub-nanometer scale. That is a legitimate milestone for 4th-generation synchrotron optics.\n\nNow the soft spots, in order of importance. First, the abstract's \"20–30x\" improvement is not what Table 3 shows: PV improvements are 10–16x, and RMS improvements range from ~17x to ~36x across the four mirrors. The word \"routinely\" overstates a best-case result—only WP#37 reaches 0.1 nm RMS; the other WP mirrors sit at 0.20–0.23 nm, and CO#7 stalls at 0.37 nm due to surface pollution. Second, the 0.1 nm value is close to the 0.05 nm RMS repeatability of the Fizeau stitching, so the true figure could be anywhere in that range. No error bars are given.\n\nThe deeper methodological concern is the untested assumption that the static flux profile, measured on stationary substrates, is the effective deposition kernel during dynamic motion. The stress-test note is right: a kernel mismatch would enter the deconvolution as a systematic bias, not as noise, and iterative convergence does not by itself prove the kernel is unbiased at the 0.1 nm scale. I weigh this as a request for additional evidence rather than a fatal flaw, because the final residuals are empirical outcomes independent of the model, and the PSD-band agreement gives indirect validation. Still, a direct open-loop test—deposit a known target profile and compare measured versus predicted thickness—would settle the question cleanly. The authors should also publish the algorithm specification or code and the raw metrology data; the paper currently references an IDL astronomy library but gives no details of the deconvolution implementation.\n\nFor whom is this paper? Anyone working on synchrotron mirror fabrication or adaptive optics for coherent X-ray sources will want to know that differential deposition has reached this level. It deserves a serious referee, not a desk reject—the engineering result is credible and useful, and the authors have a track record of careful work. My recommendation: send it to review, but ask the referee to insist on reconciling the abstract with Table 3 and to request at least one open-loop validation of the kernel equivalence if the authors can do it without much extra effort.","headline":"A credible engineering milestone—differential deposition now reaches 0.1–0.2 nm RMS on long X-ray mirrors—but the abstract overstates the improvement and the static-to-dynamic kernel assumption deserves an explicit test.","tokens_in":8506,"tokens_out":2612,"would_cite":true,"duration_ms":32260,"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":"Differential deposition of tungsten silicide films routinely reduces x-ray mirror figure errors by a factor of 20–30, reaching residuals below 0.5 nm RMS.","keywords":["x-ray optics","x-ray mirrors","differential deposition","figure correction","magnetron sputtering","thin films","interferometric stitching metrology"],"falsifier":"Deposit a known test profile in dynamic mode using a velocity profile computed from the static flux kernel, then measure the resulting thickness with the same stitched interferometry; if the residual difference from the predicted profile exceeds the stated metrology repeatability of 0.05 nm RMS, the static-kernel assumption is wrong.","tokens_in":7553,"feed_emoji":"🪞","tokens_out":10297,"duration_ms":102667,"temperature":0.7,"pith_summary":"X-ray mirrors for synchrotron beamlines must be flat to well below a nanometer to preserve coherent wavefronts, and polishing long substrates to that level is slow. This paper reports a coating-based alternative: differential deposition, in which a tungsten silicide ($\\mathrm{WSi_2}$) film with a deliberately non-uniform thickness cancels the mirror's measured height errors. The authors show on four 300 mm silicon mirrors that three corrective iterations, using successively narrower deposition apertures (24 mm, 2 mm, 1 mm), routinely reduce figure errors by a factor of 20–30, to below 0.5 nm RMS. On the best mirror the residual error reached 0.1 nm RMS and 0.5 nm peak-to-valley over a 260 mm clear aperture, down to spatial periods of 2 mm. If this result holds, differential deposition becomes a practical finishing step for long x-ray mirrors, complementing ion-beam figuring and elastic emission machining.","feed_headline":"Coating flattens x-ray mirrors to below 0.5 nm RMS","feed_subtitle":"Differential deposition cut 300 mm mirror errors 20–30x, reaching 0.1 nm RMS over a 260 mm aperture.","key_machinery":"The carrying object is the convolution of Eq. (1), $$t(x_s)=R\\int_{-S}^{S}\\frac{f(x_m-x_s)}{v(x_m)}\\,\\mathrm{d}x_m,$$ where $t(x_s)$ is the deposited thickness at substrate position $x_s$, $f(x_m-x_s)$ is the normalized static particle flux profile seen through a beam-defining aperture, $v(x_m)$ is the substrate motion speed, and $R$ is the central growth rate. Because the desired thickness is known and the speed profile is unknown, the correction is a deconvolution: the measured static flux profile is the kernel, and the algorithm solves a discretized linear system by matrix inversion, using numerical routines originally developed for astronomical image deconvolution. The second carrying element is the aperture cascade (24 mm, 2 mm, and 1 mm slits), which successively corrects long, medium, and short spatial periods, pushing the corrected spatial frequency range down to about $0.5\\,\\mathrm{mm}^{-1}$ (2 mm periods).","core_discovery":"The central claim is that differential deposition is a deterministic, repeatable figure-correction method for long x-ray mirrors at the sub-nanometer level. The paper demonstrates this on four 300 mm flat silicon mirrors coated with tungsten silicide: after an initial uniform 50 nm layer, each mirror received two corrective coatings on a large coating system and one on a compact system, with the deposition aperture narrowed from 24 mm to 2 mm to 1 mm. The measured figure errors decreased nearly exponentially with iteration, with total improvement factors between 10x and 36x depending on mirror and metric. The best residual levels were 0.1 nm RMS and 0.5 nm PV over 260 mm, with the frequency analysis showing that each aperture width suppresses height errors down to spatial periods roughly matching that width, ending at about 2 mm. The authors attribute the remaining uncorrected features to localized surface defects, short-period spikes, and hardware limits, and note that one mirror's progress was limited by surface pollution.","pith_inferences":["Inference: if the static flux profile remains the deposition kernel during motion, the same deconvolution should transfer to curved or actively bent mirrors, since only the measured height profile and kernel change.","Inference: the correction acts only along the central trace, so a two-dimensional extension would require lateral aperture shaping or a second motion axis; the paper itself notes that off-trace errors worsen.","Inference: closed-loop correction using in-situ metrology, which the authors list as future work, could remove contamination during transfer and speed up convergence."],"forward_implications":["Four 300 mm mirrors reached sub-nanometer residuals with only three corrective iterations, so differential deposition can serve as a final finishing step before multilayer coating of synchrotron mirrors.","The aperture-by-aperture cascade gives wavelength-selective correction: each narrower slit cleans up shorter spatial periods without disturbing the already corrected long periods.","Metrology repeatability below 0.05 nm RMS is sufficient to guide the correction loop, meaning off-line interferometric stitching can reliably drive figure errors to the 0.1 nm RMS scale.","The demonstrated roughly tenfold improvement in achievable RMS errors over six years, at about a doubling every 1.8 years, suggests the process still has headroom before hitting fundamental limits."],"supporting_citations":[{"why":"supplies the convolution relation between deposited thickness, static flux profile, and substrate velocity, which is the basis for the deconvolution.","marker":"[18]"},{"why":"provides the numerical deconvolution routines used to invert the system of linear equations.","marker":"[19]"},{"why":"prior demonstration of differential deposition and off-line metrology on which this work builds.","marker":"[12]"},{"why":"established that a uniform 50 nm WSi2 layer suppresses parasitic signals from buried oxide layers, letting the metrology track true figure errors.","marker":"[14]"},{"why":"describes the large coating system used for the first two corrective iterations with 24 mm and 2 mm apertures.","marker":"[16]"},{"why":"describes the compact coating system whose narrower 1 mm aperture and more precise motion enable the short-period correction.","marker":"[17]"},{"why":"documents the stitching interferometry method that reconstructs the 300 mm mirror topography with sub-nanometer accuracy.","marker":"[22]"}],"fun_headline_variants":["Differential deposition flattens x-ray mirrors to 0.1 nm RMS","Differential coating cuts mirror errors 30x to sub-nm","Sub-nm x-ray mirror figure via differential deposition","Iterative deposition corrects x-ray mirror figure to 0.1 nm RMS","X-ray mirror figure errors reduced 20-30x by differential deposition"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The correction loop assumes that the static WSi2 flux profile measured on stationary substrates is the effective deposition kernel while the substrate is moving, so the deconvolution's velocity profile deposits exactly the inverse height profile; the paper does not experimentally validate this equivalence.","fun_headline_variants_meta":{"raw":{"variants":["Differential deposition flattens x-ray mirrors to 0.1 nm RMS","Differential coating cuts mirror errors 30x to sub-nm","Sub-nm x-ray mirror figure via differential deposition","Iterative deposition corrects x-ray mirror figure to 0.1 nm RMS","X-ray mirror figure errors reduced 20-30x by differential deposition"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000811,"raw_usage":{"total_tokens":3521,"prompt_tokens":876,"completion_tokens":2645,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":492,"completion_tokens_details":{"reasoning_tokens":2550}},"tokens_in":492,"tokens_out":2645,"duration_ms":18930,"temperature":1.0,"reasoning_tokens":2550,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:43:27.643725+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Deposit a known test profile in dynamic mode using a velocity profile computed from the static flux kernel, then measure the resulting thickness with the same stitched interferometry; if the residual difference from the predicted profile exceeds the stated metrology repeatability of 0.05 nm RMS, the static-kernel assumption is wrong.","supporting_citations":[{"cited_title":"Thickness control of large area X-ray multilayers,","cited_arxiv_id":null,"evidence_quote":"supplies the convolution relation between deposited thickness, static flux profile, and substrate velocity, which is the basis for the deconvolution."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the numerical deconvolution routines used to invert the system of linear equations."},{"cited_title":"X-ray mirror figure correction by differential deposition and off-line metrology","cited_arxiv_id":null,"evidence_quote":"prior demonstration of differential deposition and off-line metrology on which this work builds."},{"cited_title":"X -ray mirrors with sub -nanometer figure errors obtained by differential deposition of thin WSi2 films,","cited_arxiv_id":null,"evidence_quote":"established that a uniform 50 nm WSi2 layer suppresses parasitic signals from buried oxide layers, letting the metrology track true figure errors."},{"cited_title":"The new ESRF multilayer deposition facility,","cited_arxiv_id":null,"evidence_quote":"describes the large coating system used for the first two corrective iterations with 24 mm and 2 mm apertures."},{"cited_title":"The new ESRF compact multilayer coating system","cited_arxiv_id":null,"evidence_quote":"describes the compact coating system whose narrower 1 mm aperture and more precise motion enable the short-period correction."},{"cited_title":"Fizeau stitching at the European Synchrotron Radiation Facility (ESRF),","cited_arxiv_id":null,"evidence_quote":"documents the stitching interferometry method that reconstructs the 300 mm mirror topography with sub-nanometer accuracy."}],"review_version":1}