{"id":"dd324653-f876-408c-aa22-004e56534ecb","arxiv_id":"2505.10752","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Sputtering Sb2S3 at low power yields near-stoichiometric films with vanishing crystalline-state extinction in the near infrared, enabling a 0.33 dB/π phase shifter at 1.55 μm.","lead":"By tuning the sputtering power and argon flow, the authors made antimony sulfide (Sb2S3) films closer to their ideal chemical makeup, which sharply reduced optical loss. This yielded a compact, selenium-free optical phase shifter with 0.33 dB loss per π phase shift at the telecom wavelength 1.55 μm.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Device-level causal claim unsubstantiated: only the optimized power is tested, and the cross-lab Fig. 11 benchmark cannot show that sputtering-power optimization enables the low-loss result.","rationale":"The paper's empirical result, 0.33 dB/π at 1550 nm, is a directly measured quantity and is not invalidated by the reader's ellipsometric-model concern. Even if the one-Tauc-Lorentz/one-Gaussian model misses weak sub-gap absorption, the waveguide loss was measured directly; the concern would only change the attribution of loss between absorption and scattering, not the headline number. Moreover, because the 20-nm Sb2S3 film is a thin cladding on a high-confinement Si waveguide, the modal confinement inside Sb2S3 is small, so even moderate material k at 1550 nm would contribute only a small fraction of the 0.013 dB/μm loss; the ellipsometric sensitivity at 1550 nm is therefore not the bottleneck for the device-level claim. The more load-bearing gap is causal: the title and abstract assert that optimizing sputtering conditions enabled the low-loss phase shifter, yet only one sputtering condition is used in the device demonstration, and the only power-dependent device comparison is a cross-laboratory literature plot (Fig. 11) with large scatter, which the authors themselves acknowledge. The blanket-film ellipsometry in Fig. 5 establishes a material-level trend, but the paper attributes most of the 10-W device loss to scattering rather than absorption; without device-level power variation, one cannot tell whether the stoichiometry optimization is what made the device low-loss. A three-condition device comparison is feasible and would settle this directly. The reader's conditional verdict remains appropriate; this concern adds a specific missing experiment rather than overturning the result.","tokens_in":12291,"tokens_out":12974,"duration_ms":140327,"concrete_test":"Fabricate the same MRR and straight-waveguide test structures with Sb2S3 deposited at 10 W, 30 W, and 50 W (Ar flow fixed at 18 sccm), keeping all other process steps identical, and measure loss per unit length, phase shift per unit length, and loss per π phase shift at 1550 nm as in Sections 3.2-3.3. If loss per π phase shift increases monotonically with sputtering power, matching the blanket-film extinction-coefficient trend in Fig. 5(d), the optimization claim is supported. If it is flat, non-monotonic, or dominated by run-to-run variation, then the low-loss result is not specifically enabled by the sputtering-power optimization.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline causal claim is that optimizing sputtering conditions (lower power, lower Ar flow) enables the 0.33 dB/π phase shifter at 1550 nm. However, all device results in Section 3 are for a single deposition condition (10 W, 18 sccm Ar). The power dependence of the figure of merit is established only from spectroscopic ellipsometry on blanket films (Section 2.2, Fig. 5), and the device-level comparison in Fig. 11 pools data from different laboratories, different tools, and different integration schemes; the authors themselves note the scatter is large. This matters because the paper argues in Section 3.2 that the 0.013 dB/μm crystalline-state loss is dominated by scattering, since ellipsometry gives k=0 at 1550 nm. If scattering dominates, the material-absorption improvement from lower sputtering power may contribute little to the total device loss, and the low loss per phase shift could instead reflect fabrication quality (roughness, sidewall coverage, film morphology) rather than the stoichiometry optimization. The title's 'Enabled by Optimizing Sputtering Conditions' is therefore not directly evidenced by the presented data. A controlled device-level comparison across sputtering powers is needed to establish the causal link.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a study of RF-magnetron-sputtered Sb2S3 films for low-loss optical phase shifters at 1.55 μm. The authors use RBS to show that the film composition approaches stoichiometric Sb2S3 as sputtering power and Ar flow are reduced, and Raman spectra corroborate the absence of Sb-Sb bonding at low power. Spectroscopic ellipsometry with a Tauc-Lorentz plus Gaussian model indicates that the extinction coefficient of crystalline Sb2S3 vanishes at shorter wavelengths for lower sputtering power, motivating a 10 W, 18 sccm deposition condition. The authors fabricate Si waveguides with a 20-nm-thick Sb2S3 cladding layer and measure a crystalline-state loss of 0.013 dB/μm and a phase shift of 0.041 π/μm, yielding a loss per π phase shift of 0.33 dB/π at 1.55 μm. The result is benchmarked against prior Sb2S3, Sb2Se3, GST, and GSTS phase shifters.","tokens_in":12561,"tokens_out":5399,"duration_ms":51517,"significance":"If the result holds, the paper provides a useful Se-free, non-volatile phase-shifter material for NIR silicon photonics and a practical recipe for sputtering near-stoichiometric Sb2S3. The manuscript has clear strengths: the RBS, Raman, and ellipsometry data are internally consistent; the headline loss per phase shift is a direct device measurement rather than a fitted extrapolation; and the comparison table and benchmark figures aggregate prior work in a transparent way. The authors also explicitly acknowledge the scatter in cross-laboratory comparisons, which is commendable. However, the central causal claim that sputtering-condition optimization enables the low-loss device is not directly evidenced, and the headline metric is extracted from very few points without uncertainty analysis. These issues are load-bearing for the paper's main claim and require revision.","major_comments":[{"comment":"The title and Abstract claim that optimizing sputtering conditions enables the low-loss phase shifter, but every device result in §3 is obtained from a single sputtering condition (10 W, 18 sccm Ar). The power dependence of the material figure of merit is established only from blanket-film ellipsometry (§2.2, Fig. 5), and the device-level benchmark in Fig. 11 pools results from different laboratories, tools, and integration schemes, with large scatter that the authors themselves acknowledge. Because the measured 0.33 dB/π could in principle reflect fabrication quality (sidewall roughness, conformality, capping-layer effects) rather than the stoichiometry improvement, the causal claim is not directly evidenced by the presented data. A device-level comparison across at least one higher sputtering power, or a substantially softened causal claim, is required.","section":"§3 and Title"},{"comment":"The headline figures of merit are extracted from very few data points with no stated uncertainties. The crystalline-state loss per unit length (0.013 dB/μm) is obtained from straight-waveguide measurements with phase-shifter lengths up to 100 μm (Fig. 9(d)), and the phase shift per unit length (0.041 π/μm) is obtained from only two micro-ring lengths, 10 μm and 20 μm (Fig. 10(d)). No error bars, numbers of repeated devices, or fit residuals are reported. Since the loss per π phase shift is derived from the ratio of these two slopes, the absence of uncertainty analysis makes it impossible to judge whether the value 0.33 dB/π is statistically distinct from neighboring values in Table 1. Please report per-length statistics, numbers of measured devices, and confidence intervals.","section":"§3.2 and §3.3"},{"comment":"The claim that crystalline Sb2S3 has exactly zero extinction coefficient at 1550 nm is a property of the chosen oscillator model (one Tauc-Lorentz and one Gaussian oscillator), not a model-independent measurement. If weak sub-gap absorption exists below the model's sensitivity, the measured 0.013 dB/μm waveguide loss would be partly absorption, weakening the assertion in §3.2 that the excess loss is dominated by scattering at surfaces and grain boundaries. A model-free validation (for example, calorimetric or photothermal measurements on the film, or a cutback series with different mode overlap) would substantiate the material-level transparency claim; alternatively, the scattering-dominance statement should be explicitly labeled as an assumption.","section":"§2.2 and §3.2"},{"comment":"The reconciliation of the measured phase shift with simulation by postulating a 12-nm Sb2S3 thickness is not securely grounded. The RBS thicknesses (13.5-14.0 nm) and ellipsometric thicknesses (18-20 nm) are for blanket films, not for the actual device in the waveguide window, and the 'agreement' with RBS is invoked only after the discrepancy is observed. This gives a circular flavor to the simulation comparison. The measured phase shift per unit length does not depend on this assumption, so the discussion should be reframed as a hypothesis and supported by a direct thickness measurement (e.g., TEM on the device cross-section) or removed.","section":"§3.3"}],"minor_comments":[{"comment":"Section numbering is duplicated: '2.2 Characterization of Optical Properties' repeats '2.2 Analysis of Composition', and §3.2 is used twice ('Measurement Setup' and 'Measurement of Loss'). Please renumber.","section":"§2.2, §3.2"},{"comment":"Figure callouts are inconsistent: the text refers to Fig. 8(a)-(c) for the straight-waveguide images and spectra, but the actual figure for these data is Fig. 9(a)-(d). Please correct the cross-references.","section":"§3.2"},{"comment":"The sentence 'The sputtering powers reported in [14], [31], and [32] are 27 sccm, 20 sccm, and 30 sccm' appears to mix units; sccm is a flow rate, not a power. Please verify the original reports and correct the wording and axis label in Fig. 11.","section":"§3.4"},{"comment":"The text says the Ar-flow-dependent atomic concentrations are 'plotted as a function of the sputtering power' in Fig. 3(b), but the scan variable is Ar flow. The caption or axis label should be corrected to avoid confusion.","section":"§2.1"},{"comment":"The ellipsometry fit residuals are not shown; a plot of the measured versus fitted Ψ and Δ spectra, or at least a discussion of the MSE values from Tables 2 and 3 in the main text, would help support the strong k=0 conclusion at 1550 nm.","section":"Appendix B"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope and the device demonstration is valuable, but the causal claim in the title and abstract outruns the data: only a single device condition is tested, and the cross-laboratory benchmark cannot substitute for a controlled comparison. The missing uncertainty analysis on the headline figure is also a serious reporting gap. I would not reject, because the gaps are fixable either by an additional device-level experiment or by appropriately softening the claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the quantitative link between sputtering power, film stoichiometry, and the crystalline-state extinction coefficient in Sb2S3. RBS shows the films get closer to stoichiometric as power and Ar flow drop, Raman corroborates the trend (Sb–Sb modes appear at high power), and ellipsometry shows the wavelength where k vanishes shifts with power. That is a clean, internally consistent story and worth having on the record. The device demonstration at 10 W gives 0.33 dB/π at 1550 nm, measured directly from waveguide loss and ring phase shift, not derived from a fitted model. That is a legitimate, useful result even if it is not a record among Sb2Se3 or other PCM phase shifters.\n\nThe soft spots are real but not fatal. First, the title says the low loss is *enabled by* optimizing sputtering conditions, yet the devices are only made at the single optimized condition. The power dependence is established on blanket films; the device-level comparison in Fig. 11 pools cross-laboratory data with different tools and integration schemes, and the scatter is large. So the causal claim is more than the data strictly support. That is an overstatement in the title, not a flaw in the measurements themselves. Second, the ellipsometric model (one Tauc–Lorentz plus one Gaussian) could miss weak sub-gap absorption, so the claim that crystalline Sb2S3 is transparent at 1550 nm rests on the model. The measured 0.013 dB/μm loss is direct, but the attribution to scattering rather than absorption is not fully proven. Third, the loss and phase-shift values come from fits without reported error bars, and the phase-shift discrepancy is closed by assuming a 12-nm film thickness rather than the 20-nm target. That thickness is justified by RBS on a different sample, but it is still a post-hoc assumption.\n\nNone of these are load-bearing enough to reject the paper. The material characterization is reproducible and the device numbers are honestly measured. The authors should either add a device-level comparison across sputtering powers or soften the causal language, and they should report error bars. For a process-optimization paper in optics, this is a solid contribution that deserves peer review, not desk rejection. I would cite it for the stoichiometry–power relationship and would bring it to reading group as an example of how to combine RBS, Raman, and ellipsometry in PCM development.","headline":"A useful process–property study that identifies sputtering power as a knob for Sb2S3 stoichiometry and shows a good (not record) phase-shifter result; the title overclaims causality, but the core material science is solid.","tokens_in":13121,"tokens_out":2452,"would_cite":true,"duration_ms":26023,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"By reducing sputtering power, Sb2S3 films become near-stoichiometric and transparent at telecom wavelengths, yielding an optical phase shifter with only 0.33 dB of loss per π phase shift at 1.55 μm.","keywords":["Sb2S3","optical phase shifter","phase change material","sputtering optimization","stoichiometry","silicon photonics","near-infrared optics","non-volatile photonics"],"falsifier":"Measure the absorption of a crystalline Sb2S3 film at 1550 nm directly—by photothermal deflection spectroscopy, or by comparing waveguide transmission against a scattering-only baseline—and check whether the extinction coefficient is zero; any nonzero value would overturn the claim that the low loss comes from transparency rather than residual absorption.","tokens_in":12118,"feed_emoji":"💡","tokens_out":10129,"duration_ms":86239,"temperature":0.7,"pith_summary":"With lower sputtering power and lower argon flow, sputtered $\\mathrm{Sb_2S_3}$ films land closer to the 40:60 Sb:S stoichiometric composition, and that compositional control determines how transparent the crystallized film is at telecom wavelengths. The paper establishes this by Rutherford backscattering composition analysis, Raman spectroscopy, and spectroscopic ellipsometry on films deposited at 10, 30, and 50 W, then confirms it in working devices: a phase shifter made from the 10 W film reaches 0.33 dB loss per $\\pi$ phase shift at 1.55 $\\mu\\mathrm{m}$, one of the lowest values reported for $\\mathrm{Sb_2S_3}$ in the near infrared. A sympathetic reader would care because this is a selenium-free, non-volatile phase-change material that could be integrated into silicon photonics without toxic elements.","feed_headline":"Sputtering trick makes Sb2S3 phase shifter lose only 0.33 dB per π","feed_subtitle":"Near-stoichiometric Sb2S3 films cut phase-shifter loss to 0.33 dB per π at 1.55 µm.","key_machinery":"The machinery is the link between sputtering conditions and film stoichiometry, read through three measurements. Rutherford backscattering gives absolute Sb and S atomic fractions and shows the Sb-rich trend with power; Raman spectroscopy gives bonding evidence, with Sb–Sb modes at 113 and 150 cm$^{-1}$ appearing as power increases; spectroscopic ellipsometry, fitted with one Tauc–Lorentz plus one Gaussian oscillator, yields the refractive index and extinction coefficient spectra. The decisive numbers are the wavelengths where the crystalline extinction coefficient vanishes: below 910 nm for 30 W and at 824 nm for 10 W, versus not at all up to 1688 nm for 50 W. Device-level loss and phase shift are extracted from straight waveguides and add-drop micro-ring resonators of different shifter lengths, giving loss per unit length and phase shift per unit length at 1.55 $\\mu\\mathrm{m}$.","core_discovery":"The central claim is that off-stoichiometry, not an intrinsic property of $\\mathrm{Sb_2S_3}$, is the main source of the residual near-infrared extinction seen in crystalline $\\mathrm{Sb_2S_3}$ films. RBS shows films sputtered at 30 and 50 W are Sb-rich (Sb:S = 43.2:56.8 and 43.8:56.2 at.%) while the 10 W film is slightly S-rich (38.8:61.2), and Raman spectra reveal Sb–Sb bonding modes appearing at higher power. The 10 W crystalline film's extinction coefficient reaches zero at 824 nm, so it is transparent at 1550 nm, whereas the 50 W film remains absorbing across the whole measured range to 1688 nm. In fabricated silicon waveguides, the optimized film gives a crystalline-state loss of 0.013 dB/$\\mu\\mathrm{m}$ and a phase shift of 0.041 $\\pi/\\mu\\mathrm{m}$, combining to 0.33 dB/$\\pi$ at 1550 nm.","pith_inferences":["If off-stoichiometry is the common cause of residual near-infrared absorption in chalcogenide phase-change films, the same power-and-flow optimization should transfer to $\\mathrm{Sb_2Se_3}$ and GSTS films, potentially lowering their losses without changing material chemistry.","The measured phase shift (0.041 $\\pi/\\mu\\mathrm{m}$) being smaller than the simulated value (0.067 $\\pi/\\mu\\mathrm{m}$) suggests the active film is roughly 12 nm rather than 20 nm; a thicker film would shorten the device but trade in some loss, a trade-off the paper leaves unquantified.","A direct absorption measurement on the crystalline film—rather than an ellipsometric model—would separate scattering from sub-gap absorption and determine how much headroom remains in the 0.33 dB/$\\pi$ figure."],"forward_implications":["A $\\mathrm{Sb_2S_3}$ phase shifter with 0.33 dB/$\\pi$ at 1550 nm is competitive with the best selenium-containing phase shifters while avoiding selenium, so it becomes a realistic non-volatile building block for silicon photonic circuits.","Because crystalline $\\mathrm{Sb_2S_3}$ is effectively transparent at the operating wavelength, the device performs near-pure phase modulation: switching changes refractive index without modulating intensity.","The 10 W film's strong visible-light absorption makes it suitable for laser-writing and amorphization, so the same low-loss film can act as a rewritable platform for arbitrary optical components.","Adopting the optimized sputtering condition (10 W, 18 sccm) is a directly transferable recipe for other groups making $\\mathrm{Sb_2S_3}$ photonics, which should reduce the spread of reported losses."],"supporting_citations":[{"why":"Identifies Sb2S3 and Sb2Se3 as ultralow-loss wide-gap phase-change materials and reports that the band gap varies with sputtering power and pressure, motivating the composition study.","marker":"[13]"},{"why":"Supplies the amorphous and crystalline band gaps (about 2.1 eV and 1.7 eV) used to expect zero interband absorption at 1550 nm.","marker":"[27]"},{"why":"Reports an earlier Sb2S3 phase shifter with 5.93 dB/π loss at 1550 nm, the main performance baseline this work improves on.","marker":"[14]"},{"why":"Provides Sb2S3 and Sb2Se3 phase-shifter loss and phase-shift data at 1310 and 1550 nm, used as comparison benchmarks.","marker":"[15]"},{"why":"Reports a non-volatile electrically programmable Sb2S3 device whose loss per phase shift is benchmarked in Table 1.","marker":"[17]"},{"why":"Reports Sb2S3 on a 300-mm silicon photonics platform and serves as another near-infrared loss benchmark.","marker":"[18]"},{"why":"Supplies equations (1) and (2) used to estimate loss per phase shift from ellipsometry and proposes the selenium-free GSTS material concept.","marker":"[12]"},{"why":"Assigns the Raman modes used here to identify Sb–Sb bonding that signals Sb-rich composition.","marker":"[29]"},{"why":"Reports the sputtering power used for its Sb2S3 film, serving as a benchmark data point in Fig. 11.","marker":"[31]"},{"why":"Reports the sputtering power used for its Sb2S3 devices, serving as another benchmark data point in Fig. 11.","marker":"[32]"}],"fun_headline_variants":["Sputtering tune-up yields Sb2S3 phase shifter at 0.33 dB/π","Stoichiometric Sb2S3 films enable low-loss phase shifting at 1550 nm","Low-power sputtering cuts Sb2S3 phase shifter loss to 0.33 dB/π","Optimized sputtering gives Sb2S3 phase shifter with 0.33 dB/π loss","Sb2S3 phase shifter hits 0.33 dB/π via sputtering optimization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the ellipsometric fitting model—one Tauc–Lorentz oscillator plus one Gaussian oscillator—correctly determines that crystalline Sb2S3 has zero absorption at 1550 nm; if weak sub-gap absorption is hiding below the model's sensitivity, part of the measured 0.013 dB/μm loss is absorption, not scattering, and the material is not truly transparent there.","fun_headline_variants_meta":{"raw":{"variants":["Sputtering tune-up yields Sb2S3 phase shifter at 0.33 dB/π","Stoichiometric Sb2S3 films enable low-loss phase shifting at 1550 nm","Low-power sputtering cuts Sb2S3 phase shifter loss to 0.33 dB/π","Optimized sputtering gives Sb2S3 phase shifter with 0.33 dB/π loss","Sb2S3 phase shifter hits 0.33 dB/π via sputtering optimization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000673,"raw_usage":{"total_tokens":3061,"prompt_tokens":936,"completion_tokens":2125,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":1998}},"tokens_in":552,"tokens_out":2125,"duration_ms":12974,"temperature":1.0,"reasoning_tokens":1998,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:03:29.105774+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the absorption of a crystalline Sb2S3 film at 1550 nm directly—by photothermal deflection spectroscopy, or by comparing waveguide transmission against a scattering-only baseline—and check whether the extinction coefficient is zero; any nonzero value would overturn the claim that the low loss comes from transparency rather than residual absorption.","supporting_citations":[{"cited_title":"A new family of ultralow loss reversible phase‐change materials for photonic integrated circuits: Sb 2 S 3 and Sb 2 Se 3,","cited_arxiv_id":null,"evidence_quote":"Identifies Sb2S3 and Sb2Se3 as ultralow-loss wide-gap phase-change materials and reports that the band gap varies with sputtering power and pressure, motivating the composition study."},{"cited_title":"Wide bandgap phase change material tuned visible photonics,","cited_arxiv_id":null,"evidence_quote":"Supplies the amorphous and crystalline band gaps (about 2.1 eV and 1.7 eV) used to expect zero interband absorption at 1550 nm."},{"cited_title":"Non‐volatile reconfigurable integrated photonics enabled by broadband low‐loss phase change material,","cited_arxiv_id":null,"evidence_quote":"Reports an earlier Sb2S3 phase shifter with 5.93 dB/π loss at 1550 nm, the main performance baseline this work improves on."},{"cited_title":"Towards low loss non-volatile phase change materials in mid index waveguides,","cited_arxiv_id":null,"evidence_quote":"Provides Sb2S3 and Sb2Se3 phase-shifter loss and phase-shift data at 1310 and 1550 nm, used as comparison benchmarks."},{"cited_title":"Non-volatile electrically programmable integrated photonics with a 5 -bit operation,","cited_arxiv_id":null,"evidence_quote":"Reports a non-volatile electrically programmable Sb2S3 device whose loss per phase shift is benchmarked in Table 1."},{"cited_title":"Deterministic quasi - continuous tuning of phase-change material integrated on a high-volume 300-mm silicon photonics platform,","cited_arxiv_id":null,"evidence_quote":"Reports Sb2S3 on a 300-mm silicon photonics platform and serves as another near-infrared loss benchmark."},{"cited_title":"Proposal of low -loss non-volatile mid-infrared optical phase shifter based on Ge 2Sb2Te3S2,","cited_arxiv_id":null,"evidence_quote":"Supplies equations (1) and (2) used to estimate loss per phase shift from ellipsometry and proposes the selenium-free GSTS material concept."},{"cited_title":"In situ analysis of the crystallization process of Sb 2 S 3 thin films by Raman scattering and X -ray diffraction,","cited_arxiv_id":null,"evidence_quote":"Assigns the Raman modes used here to identify Sb–Sb bonding that signals Sb-rich composition."},{"cited_title":"Comparison and analysis of phase change materials-based reconfigurable silicon photonic directional couplers,","cited_arxiv_id":null,"evidence_quote":"Reports the sputtering power used for its Sb2S3 film, serving as a benchmark data point in Fig. 11."},{"cited_title":"Capping layer effects on Sb2S3-based reconfigurable photonic devices,","cited_arxiv_id":null,"evidence_quote":"Reports the sputtering power used for its Sb2S3 devices, serving as another benchmark data point in Fig. 11."}],"review_version":1}