{"id":"a54b9314-a846-4c6b-86c7-dc461228ae1c","arxiv_id":"2607.17905","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"CO2-laser-annealed Ce:YIG on a YIG seed layer yields 27.1 dB isolation in a SiN Mach-Zehnder isolator at 1552.7 nm, but with higher loss than RTA-grown films.","lead":"Researchers crystallized cerium-substituted yttrium iron garnet (Ce:YIG) on silicon and silicon-nitride photonic chips using a localized CO2 laser anneal instead of a full-chip high-temperature bake, then built a one-way light valve (optical isolator) that reached 27.1 dB isolation at 1552.7 nm. A smart generalist might read this because it is a step toward putting magnetic optical components on silicon photonic circuits without violating the low-temperature limits of chip ma","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Ce:YIG propagation loss of 188.2 dB/cm rests on Eq. (1) with unquantified uncertainties in confinement factors, assumed SiN loss, and unchanged YIG loss; direct cutback measurement should settle it.","rationale":"The central claim includes a specific propagation loss of 188.2±16.5 dB/cm. Although the device demonstration (isolation ratio 27.1 dB, insertion loss 10.1 dB) is a direct measurement and survives even if the extracted loss is inaccurate, the paper's quantitative performance comparison to RTA and its assessment of the laser annealing route rest on this number. The extraction via Eq. (1) is a three-term model with no propagated uncertainties: the confinement factors come from a simulation with no error estimate, the YIG loss is taken from a pre-deposition measurement assumed unchanged, and the SiN loss is taken from a previous publication. Given the small Ce:YIG confinement (11.37%), the amplified sensitivity of α_Ce:YIG to these assumptions makes the quoted value fragile. The proposed cutback measurement is a standard, model-independent method that would directly measure the modal loss in the Ce:YIG-covered waveguide, thereby validating or refuting the Eq. (1) result. We agree with the reader's identified weakest assumption and see no need to change the CONDITIONAL verdict; the paper should be accepted only after this loss measurement is provided or the claims are softened. Credit is due for the direct isolation ratio and insertion loss measurements, which support the existence proof regardless of the exact loss value.","tokens_in":14376,"tokens_out":8508,"duration_ms":75387,"concrete_test":"Fabricate a set of identical SiN waveguides with the same Ce:YIG/YIG stack but with Ce:YIG-covered lengths of 0.5, 1.0, 2.0, and 4.0 mm, using the same laser annealing process. Measure TM insertion loss for each (normalized to reference waveguides without Ce:YIG) and plot loss vs length. The slope yields the total propagation loss directly, without using Eq. (1). Compare the derived Ce:YIG contribution to the reported 188.2 dB/cm (accounting for known YIG and SiN losses). If the derived α_Ce:YIG differs by more than ~20%, the Eq. (1) extraction is unreliable. Additionally, re-measure the YIG seed loss after the full Ce:YIG anneal to check whether it remains 75.9 dB/cm.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3.3, Eq. (1) defines the waveguide loss as a weighted sum of Ce:YIG, YIG, and SiN losses. The extracted Ce:YIG loss (188.2±16.5 dB/cm) is a headline quantitative result, but it inherits three unverified inputs: (i) COMSOL confinement factors (11.37%, 6.17%, 65.75%) with no error bars; (ii) a YIG loss (75.9±5.4 dB/cm) measured before Ce:YIG deposition, which may change after the second laser anneal (e.g., interdiffusion or further crystallization); (iii) an assumed SiN loss of 0.5 dB/cm from ref. 50, not measured on this chip. Because the Ce:YIG confinement is only ~11%, a small systematic error in the simulated mode profile (e.g., ±20% relative) shifts the extracted α_Ce:YIG by tens of dB/cm, potentially moving it from 'inferior to RTA' (188 vs 79 dB/cm) to comparable or worse. Additionally, ellipsometry in §3.2 reports k=0 at 1550 nm for the optimal film; if the material is truly transparent, the 188 dB/cm must arise entirely from scattering/radiation, making the additive absorption-weighted model even less reliable. The isolation ratio and insertion loss are direct measurements, so the device demonstration survives; but the quantitative loss claim and the comparison to RTA are not robust as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a CO2-laser-annealing route for crystallizing 120 nm Ce:YIG films on SiO2/Si substrates and SiN waveguides, using a pre-crystallized YIG seed layer to stabilize the garnet phase. It systematically studies laser power and oxygen partial pressure and reports phase-pure, crack-free films with Ms = 146.3 emu/cm3, saturation Faraday rotation of -6573±375 deg/cm at 1200 nm and -2544±310 deg/cm at 1550 nm, comparable to RTA-grown films. On SiN waveguides, the authors fabricate a Mach-Zehnder interferometer isolator and report an NRPS of 5.78 rad/cm, Faraday rotation -2317.7±94.8 deg/cm, propagation loss 188.2±16.5 dB/cm, isolation ratio 27.1 dB, and insertion loss 10.1 dB at 1552.7 nm. They also demonstrate spatial confinement of the >400 °C thermal region to about 2.3 mm with a focused laser spot.","tokens_in":14752,"tokens_out":4754,"duration_ms":45803,"significance":"If confirmed, this would be an important existence proof: localized laser annealing of Ce:YIG on a silicon-compatible platform, without full-wafer high-temperature annealing, yielding a working nonreciprocal MZI isolator. The paper's direct measurements — XRD, EBSD, XPS, VSM, free-space Faraday rotation, and the isolation/insertion-loss spectra — are substantial strengths. The main quantitative claim that needs scrutiny is the extracted Ce:YIG propagation loss, which is not directly measured and rests on several assumptions. The device demonstration and the free-space MO properties are nevertheless valuable and largely independent of that extraction.","major_comments":[{"comment":"The headline α_Ce:YIG = 188.2±16.5 dB/cm is load-bearing for the comparison to RTA films (79±5.0 dB/cm), but it is not directly measured. Equation (1) assumes the total waveguide loss is an additive weighted sum, using (i) COMSOL confinement factors 11.37%, 6.17%, 65.75% with no uncertainty; (ii) α_SiN = 0.5 dB/cm taken from ref. 50 rather than measured on this chip; and (iii) α_YIG = 75.9±5.4 dB/cm measured before Ce:YIG deposition, which may change after the second laser anneal. Because Γ_Ce:YIG is only about 11%, a modest systematic error in the simulated mode profile or in α_SiN shifts the extracted Ce:YIG loss by tens of dB/cm — enough to move the result from 'inferior to RTA' to 'comparable to RTA.' Please provide a direct cutback measurement or, at minimum, a sensitivity analysis with error propagation, and justify the assumption that the YIG seed loss is unchanged after Ce:YIG an","section":"§3.3, Eq. (1)"},{"comment":"There is an internal inconsistency between the ellipsometry result and the loss extraction. Section 3.2 and Fig. 5(f) report an extinction coefficient k = 0 at 1550 nm for the optimal LA-Ce:YIG film, with the text saying the extinction coefficient 'could not be accurately measured by ellipsometry, indicating a relatively low loss.' Yet §3.3 attributes 188.2 dB/cm of waveguide loss to Ce:YIG via Eq. (1). If the material is truly transparent at 1550 nm, the waveguide loss must be dominated by scattering or radiation, and the additive absorption-weighted model in Eq. (1) is not appropriate. If instead the ellipsometric k is unreliable, that should be stated explicitly. This reconciliation is needed for the quantitative loss claim to be credible.","section":"§3.2 vs §3.3"}],"minor_comments":[{"comment":"Grammar/clarity: 'Figures 5(f) shows' should be 'Figure 5(f) shows'; 'extinction ratio' should be 'extinction coefficient' where referring to k.","section":"§3.2, Fig. 5(f)"},{"comment":"The phrase 'propagation loss of 188.2±16.5 dB/cm' should be qualified as an extracted/model-dependent value, not a directly measured material loss.","section":"Abstract and §3.3"},{"comment":"Please clarify how the 400 °C and 850 °C thresholds are calibrated on the infrared camera image; the spatial confinement claim depends on this calibration.","section":"Fig. 6(a)"},{"comment":"Reference formatting is inconsistent (e.g., refs. 31 and 50 mix journal, title, and volume information in nonstandard ways). A careful cleanup would improve readability.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The direct device demonstration — 27.1 dB isolation and 10.1 dB insertion loss — is solid and publishable in principle. The main risk is that the headline Ce:YIG loss and the 'comparable to RTA' framing are built on an unvalidated additive model. If the authors can add a direct loss measurement or substantially caveat the extraction, I would support acceptance after revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe genuinely new thing here is a working MZI isolator made from CO2-laser-annealed Ce:YIG on a YIG seed layer on SiN. That is the first integrated nonreciprocal device from laser-annealed Ce:YIG I know of, and the 27.1 dB isolation and 10.1 dB insertion loss are direct measurements. The free-space Faraday rotation is close to what RTA gives, and the XRD/EBSD/XPS evidence for phase purity is solid. So the existence proof holds.\n\nThe soft spot is the 188.2 dB/cm propagation loss for Ce:YIG. It is not measured; it is extracted from the MZI loss using Eq. (1) with COMSOL confinement factors (11.37% for Ce:YIG) and an assumed SiN loss of 0.5 dB/cm from a previous paper. If the simulated mode profile is off by 20%, the extracted Ce:YIG loss swings by tens of dB/cm. They also report k = 0 at 1550 nm from ellipsometry and then attribute 188 dB/cm of loss to Ce:YIG. That is not a hard contradiction if the loss is scattering, but it means the additive confinement-weighted model is not the right tool unless they say so explicitly. They do not propagate uncertainty in the confinement factors or the assumed SiN loss, so the ±16.5 dB/cm is only the fit uncertainty, not the systematic error.\n\nThe paper does compare itself to RTA Ce:YIG and openly states the device is inferior (79 dB/cm, -2500 deg/cm). That is honest and appropriate for a process demonstration. The abstract, however, presents the 188.2 dB/cm without the model-dependent caveat, which is misleading.\n\nCitation-wise, they cite the prior vacuum laser annealing of Ce:YIG (ref 30) and do not overclaim novelty relative to it. The novelty is the device integration, and that is real.\n\nOverall: the central claim—localized laser annealing works for Ce:YIG on SiN and yields a functioning isolator—is supported. The quantitative loss claim needs better support. A direct cutback or ring-resonator measurement would settle it. I would send this to peer review; the process is of interest to people working on BEOL-compatible magneto-optic integration, and the issues are fixable with a few extra experiments and a more careful error budget.","headline":"First laser-annealed Ce:YIG isolator on SiN works, but the headline propagation loss is an inference with unquantified assumptions.","tokens_in":15298,"tokens_out":2431,"would_cite":true,"duration_ms":23228,"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":"The paper establishes that CO2 laser annealing, using a laser-crystallized YIG seed layer, can produce phase-pure Ce:YIG films on silicon-based photonic chips, and that a SiN Mach-Zehnder isolator made this way achieves 27.1 dB isolation at","keywords":["Ce:YIG","CO2 laser annealing","magneto-optical garnet","YIG seed layer","optical isolator","silicon nitride photonics","nonreciprocal photonics","Faraday rotation"],"falsifier":"Directly measure the propagation loss of a laser-annealed Ce:YIG/YIG/SiN waveguide without assuming the SiN loss—for example, by cutback measurements across several waveguide lengths or by ring-resonator linewidth fitting—and compare the result with the extracted 188±16.5 dB/cm. A significant disagreement would show that the confinement-weighted extraction in Eq. (1) is the source of error.","tokens_in":14264,"feed_emoji":"🛡️","tokens_out":5678,"duration_ms":56334,"temperature":0.7,"pith_summary":"This paper tries to establish a backend-compatible way to put magneto-optical garnet films on silicon photonic chips: instead of heating the whole wafer above 800°C, a CO2 laser is used to anneal a thin Ce:YIG film locally, with a pre-crystallized YIG seed layer stabilizing the garnet phase. The authors show that the laser-annealed films achieve Faraday rotation and magnetization close to those of rapid-thermal-annealed films, and they demonstrate a working SiN Mach-Zehnder isolator with 27.1 dB isolation and 10.1 dB insertion loss. If true, this is an existence proof that nonreciprocal photonic devices can be made on silicon without a full-wafer high-temperature anneal, which matters for integrating isolators and circulators into photonic circuits. The main caveat is that the headline Ce:YIG propagation loss of 188 dB/cm is extracted indirectly from device loss using simulated mode confinement rather than measured directly.","feed_headline":"Laser-annealed Ce:YIG makes a working on-chip optical isolator","feed_subtitle":"A CO2 laser plus a YIG seed layer crystallizes the film locally, yielding 27.1 dB isolation at 1552.7 nm.","key_machinery":"The load-bearing mechanism is a two-step CO2 laser annealing process: a 50 nm YIG seed layer is first laser-crystallized, then a 120 nm Ce:YIG film deposited on it is annealed with the same 10.6 µm laser. The seed layer acts as a structural template that stabilizes the garnet phase and suppresses cracks and secondary phases. The laser is absorbed mainly by the SiO2 layer, creating a localized high-temperature zone while keeping the surrounding chip below 400°C. The nonreciprocal device is a Mach-Zehnder interferometer in which the Ce:YIG-loaded arm provides a nonreciprocal phase shift; the quoted Ce:YIG propagation loss is extracted from the measured total waveguide loss using simulated moda","core_discovery":"On the paper's own terms, the discovery is that a sputtered Ce:YIG film can be crystallized locally on silicon-based photonic platforms by CO2 laser annealing when it is grown on a laser-crystallized YIG seed layer. The resulting films reach roughly 97.5% garnet phase purity, are crack-free, and show Faraday rotations of about -6573 deg/cm at 1200 nm and -2544 deg/cm at 1550 nm, comparable to rapid-thermal-annealed references. On a SiN Mach-Zehnder isolator, the laser-annealed Ce:YIG produces a nonreciprocal phase shift of 5.78 rad/cm, corresponding to a Faraday rotation of about -2317.7 deg/cm, and a 27.1 dB one-way isolation ratio at 1552.7 nm, with an on-chip insertion loss of 10.1 dB. Th","pith_inferences":["Because the 188 dB/cm Ce:YIG loss is not directly measured, the true value could be lower if the assumed SiN loss of 0.5 dB/cm is overestimated; a direct cutback or ring-resonator measurement would settle the comparison with RTA films.","The seed-layer approach likely generalizes beyond Ce:YIG: any garnet that struggles to nucleate on amorphous silicon-compatible substrates could be grown on a laser-crystallized YIG template, removing the need for lattice-matched substrates.","The demonstrated thermal confinement suggests that with a smaller spot or a visible-wavelength laser, crystallization could be confined to a single circuit block, enabling per-device annealing in a chip-scale manufacturing flow.","The remaining loss gap relative to RTA films (188 vs 79 dB/cm) points to oxygen stoichiometry and Ce³⁺ oxidation as the main lever; suppressing Ce³⁺ oxidation during laser annealing could close most of that gap."],"forward_implications":["Laser annealing could replace full-wafer high-temperature annealing for Ce:YIG crystallization, enabling monolithic magneto-optical isolators, circulators, and switches in back-end silicon photonic processing.","The YIG seed layer is sufficient to suppress secondary phases and cracks in Ce:YIG, suggesting the same bilayer strategy may work for other garnet compositions that are hard to crystallize on amorphous substrates.","The demonstrated MZI isolator is, by the authors' account, the first integrated nonreciprocal device based on laser-annealed Ce:YIG, with an isolation ratio of 27.1 dB at 1552.7 nm.","Thermal localization can be improved from a ~2.3 mm hot zone toward ~200 µm by shrinking the laser spot or using shorter-wavelength lasers, so only the device area would be exposed to high temperature.","The measured Faraday rotation of about -2317.7 deg/cm is close to the rapid-thermal-annealed reference of -2500 deg/cm, indicating the localized route can approach conventional annealing performance."],"fun_headline_variants":["CO2 laser crystallizes Ce:YIG on silicon photonics","Laser-annealed Ce:YIG hits 27.1 dB on-chip isolation","Seed layer enables laser-grown garnet on silicon nitride","Ce:YIG isolator on SiN via CO2 laser annealing"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing assumption is that the device's total measured loss is exactly the sum of Ce:YIG, YIG, and SiN losses weighted by simulated confinement factors, with the SiN loss fixed at 0.5 dB/cm from earlier work; if the simulated mode profile, the assumed SiN loss, or the additivity assumption is wrong, the headline Ce:YIG propagation loss and the comparison to RTA films change.","fun_headline_variants_meta":{"raw":{"variants":["CO2 laser crystallizes Ce:YIG on silicon photonics","Laser-annealed Ce:YIG hits 27.1 dB on-chip isolation","Seed layer enables laser-grown garnet on silicon nitride","Ce:YIG isolator on SiN via CO2 laser annealing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000611,"raw_usage":{"total_tokens":2740,"prompt_tokens":865,"completion_tokens":1875,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":1797}},"tokens_in":609,"tokens_out":1875,"duration_ms":12375,"temperature":1.0,"reasoning_tokens":1797,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T16:38:11.513506+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Directly measure the propagation loss of a laser-annealed Ce:YIG/YIG/SiN waveguide without assuming the SiN loss—for example, by cutback measurements across several waveguide lengths or by ring-resonator linewidth fitting—and compare the result with the extracted 188±16.5 dB/cm. A significant disagreement would show that the confinement-weighted extraction in Eq. (1) is the source of error.","supporting_citations":[],"review_version":1}