{"id":"e3319657-03f1-4214-ad20-65891986a422","arxiv_id":"2505.08668","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Inverse-designed 3 µm transverse-mode beamsplitters achieve Hong-Ou-Mandel visibility up to 99.56±0.64%, averaging 99.38±0.41% across three copies.","lead":"Researchers built tiny silicon chips that mix two different transverse shapes, or modes, of light inside a 3-by-3-micron area. The chips show near-perfect quantum interference between photon pairs, with up to 99.56% visibility, pointing toward compact high-dimensional photonic quantum devices.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unreported multiplexer/demultiplexer crosstalk leaves attribution of the near-unity HOM dip to the mode beamsplitter partially underdetermined.","rationale":"The paper is a well-executed experimental demonstration with source characterization, multiple device copies, and agreement between the measured HOM visibility and the model in Eq. (5). The high visibility itself is strong evidence: same-mode crosstalk in the multiplexer would dilute the dip, so the 99.56% value indirectly bounds the input-state contamination. However, the specific attribution of the dip to the inverse-designed mode beamsplitter rather than to the combined multiplexer+beamsplitter+demultiplexer system is not fully established without reporting the crosstalk of the multiplexing stages. Because Eq. (5) describes the total device scattering matrix, matching the beamsplitter-only eta_eff curve is suggestive but not decisive, especially near eta_eff = 0.5 where the predicted visibility is nearly flat. The missing extinction-ratio and higher-order-mode data are therefore a legitimate condition, matching the reader's weakest assumption. No more severe objection surfaces: the derivation of Eq. (5) is correct, the loss model is appropriate, and the reported uncertainties are consistent with the stated integration times. The verdict should remain CONDITIONAL, unchanged from the reader.","tokens_in":10279,"tokens_out":38474,"duration_ms":380115,"concrete_test":"Fabricate test structures of the multiplexer/demultiplexer alone and measure the classical crosstalk matrix (|S32|^2 and |S41|^2 from TE0/TE1 inputs to the two output waveguides) at 1548 nm with a mode-resolved setup; report the extinction ratio and the power in TE2/radiation modes at the output of the multiplexer. If the extinction ratio exceeds 20 dB and higher-order content is below 1%, the assumption of one photon per mode is validated and the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central result is a HOM visibility of up to 99.56% measured on a chip containing a mode multiplexer, a 3 by 3 micron inverse-designed mode beamsplitter, and a demultiplexer. The theory curve in Fig. 7(b) is computed from Eq. (5) using the beamsplitter's classical effective splitting ratio eta_eff, yet Eq. (5) applies equally to the full input-fibre-to-output-fibre scattering matrix. The main text does not report the multiplexer/demultiplexer extinction ratios (the S32 and S41 crosstalk terms in Eq. (8)) or the higher-order-mode content at the beamsplitter input. If crosstalk in either mode-multiplexing stage is non-negligible, the two-photon input state is not the assumed |TE0,TE1> state, and the near-unity visibility could reflect the combined multiplexer+beamsplitter transformation rather than the mode beamsplitter in isolation. Same-mode crosstalk would dilute the dip, so the high visibility indirectly bounds crosstalk, but that bound is not reported and the attribution of the interference to the inverse-designed beamsplitter is therefore not fully pinned down by the presented data.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript reports the design, fabrication, and quantum characterization of inverse-designed transverse-mode beamsplitters in a 3 µm × 3 µm footprint on a silicon-on-insulator platform. The device under test is a chain consisting of grating couplers, an inverse-designed two-mode multiplexer, the mode beamsplitter, and a demultiplexer. The authors measure Hong-Ou-Mandel interference between TE0 and TE1 spatial modes, reporting a maximum visibility of 99.56 ± 0.64% from a single device, an average visibility across three identical copies of Design B of 99.38 ± 0.41%, and visibilities above 98% for all three inverse-designed beamsplitter designs. The theoretical benchmark is obtained from Eq. (5) using the classically measured effective splitting ratio η_eff and an independently measured source visibility of 99.57 ± 0.30%. Fabrication-bias simulations are used to argue robustness of the designs.","tokens_in":10509,"tokens_out":10829,"duration_ms":106863,"significance":"If the central attribution is fully supported, this is a valuable demonstration that inverse-designed ultra-compact multimode components can be operated in the quantum regime with near-unity interference visibility. The paper has several genuine strengths: the source visibility is measured independently, the HOM prediction is not obtained by fitting the HOM data but from the separately characterized η_eff, multiple device copies are measured, and fabrication-tolerance simulations are included. The 3 µm × 3 µm footprint is substantially smaller than previous transverse-mode beamsplitter implementations, and the reproducibility across nominally identical devices is a useful practical result for integrated quantum photonics. The main caveat is that the experiment characterizes the whole multiplexer-beamsplitter-demultiplexer chain, and the manuscript does not yet establish that the near-unity visibility is attributable to the inverse-designed beamsplitter in isolation rather than to the combined response of the full device.","major_comments":[{"comment":"The central attribution of the measured near-unity HOM visibility to the inverse-designed mode beamsplitter is not fully supported without characterizing the surrounding multiplexer/demultiplexer. The device under test is a chain of grating couplers, a two-mode multiplexer, the beamsplitter, and a demultiplexer, but the main text reports no measured extinction ratios for the multiplexer/demultiplexer (the |S32|^2 and |S42|^2 terms in Eq. (8) are only optimization targets), nor the higher-order-mode content at the beamsplitter input. Since Eq. (5) models a two-port beamsplitter acting on |TE0,TE1>, any crosstalk amplitude in the mode multiplexers changes the input state and hence the visibility; the measured ≈99.5% visibility indirectly bounds this crosstalk, but the bound is not quantified. Please report the measured mux/demux extinction ratios and demonstrate (analytically or numerically) that the residual crosstalk changes the predicted V_HOM by less than the reported uncertainty.","section":"Section 4, Eq. (5), Fig. 7"},{"comment":"The theoretical benchmark in Fig. 7(b) is computed from η_eff 'based on classical characterisation' (Supplement 9.3), but the main text does not specify whether η_eff was measured on the full device or simulated for the beamsplitter alone, nor the uncertainty in η_eff and its propagation through Eq. (5). In addition, the prediction assumes cosα = −1; the text says the optimized designs are 'consistent with energy conservation' but does not report a measured or simulated value of α for each device. Please clarify these points and give the propagated uncertainty of the predicted visibility so that the agreement in Fig. 7(b) can be assessed quantitatively.","section":"Section 4, Fig. 7(b), Section 5"},{"comment":"The reproducibility claim rests on the average visibility 99.38 ± 0.41% across 'three identical devices,' but the text does not state whether this is the standard deviation or standard error of the three fitted visibilities, and it does not specify how many copies of Designs A and C were measured. Because reproducibility is a central claim of the paper, please define the averaging procedure and report the per-design device counts and per-copy uncertainties.","section":"Section 4, Fig. 7(b)"}],"minor_comments":[{"comment":"The statement that a perfect SPDC source would give visibilities 'up to 99.95%' with the best device appears inconsistent with the quoted η_eff = 48.77% for Design B: Eq. (5) yields approximately 99.88% for that value. Please reconcile the numbers or rephrase the claim.","section":"Section 5"},{"comment":"The main text reports the source visibility of 99.57 ± 0.30% but does not show the corresponding off-chip HOM dip or its raw data; including this measurement in the main text or supplement would make the source-limited comparison in Fig. 7(b) more transparent.","section":"Section 4, Fig. 7"},{"comment":"The claim that the device is 'the smallest transverse mode beamsplitters for 1550 nm photons to date' is supported only by a rough comparison with the 60 µm grating-based device in Ref. [25]; please provide a broader comparison or qualify the claim as applying to the designs considered here.","section":"Section 1, Abstract"},{"comment":"The phrase 'the coefficient amplitudes should be symmetric' in the context of a lossless beamsplitter should be stated more precisely as |t1| = |t2| and |r1| = |r2| under the chosen phase convention, to avoid confusion with the separate notion of an unbalanced splitting ratio.","section":"Section 3, Eq. (2)"},{"comment":"Given that the central quantitative result is a visibility of 99.56 ± 0.64%, the paper would benefit from including the raw coincidence-count data and fit residuals as supplementary material rather than only providing them on request.","section":"Data availability"}],"recommendation":"major_revision","confidential_remarks":"The key unresolved issue is the measured multiplexer/demultiplexer crosstalk and the corresponding attribution of the near-unity HOM dip to the beamsplitter alone. If Supplement 9.3 already contains the relevant extinction-ratio measurements, the revision may be straightforward; otherwise, the authors should add those measurements or explicitly reframe the claim as applying to the full inverse-designed device chain."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Hi — quick read of arXiv:2505.08668. Short version: it's a clean, well-executed demonstration that inverse-designed mode beamsplitters deliver near-unity HOM visibility between TE0 and TE1 transverse modes in a 3x3 µm footprint. The repeatability across three copies of the same design is the strongest part. I'd send it to peer review.\n\nWhat's genuinely new: previous transverse-mode HOM used a Bragg grating [25]; here inverse design gives a much smaller device with comparable or better visibility. The test of theory against classical characterization is honest: they measure η_eff from classical transmission and plug into Eq. (5), so the predicted curve isn't fitted to the quantum data. Fabrication-bias simulations are sensible. Source visibility independently measured at 99.57±0.30% — good practice.\n\nWhere the paper is soft: the stress-test note about multiplexer/demultiplexer crosstalk is on target, but the sting is less than the note suggests. Same-mode leakage (say TE0 leaking into the TE1 input port) would dilute the HOM dip, so a 99.5% visibility already bounds that crosstalk to well below 1%. The more legitimate concern is that the paper never reports the extinction ratios or higher-order mode content, so the attribution to the beamsplitter itself is indirect. A reviewer should ask for those numbers in the main text or supplement. Also the 'smallest to date' claim needs a fair comparison with other compact mode beamsplitters — they mention the grating length but not a systematic survey. No raw data or device files are included; the data availability line says 'upon reasonable request,' which is fine for a letter but blocks exact reproduction.\n\nOverall the central argument holds up. This is a solid engineering advance, not a new physics principle. The right audience is experimental integrated photonics researchers; for them it's a useful data point. I'd accept for peer review, with requests for the missing crosstalk numbers and data release. Bring it to reading group if your group thinks about mode-encoded quantum photonics — otherwise read the references and move on.","headline":"A solid, reproducible demonstration of near-unity HOM visibility in inverse-designed mode beamsplitters; the missing crosstalk numbers are a real but minor gap.","tokens_in":11084,"tokens_out":3238,"would_cite":true,"duration_ms":29639,"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":"An inverse-designed 3 µm by 3 µm mode beamsplitter achieves Hong-Ou-Mandel visibility up to 99.56 percent, with every design above 98 percent, showing that transverse spatial modes can support near-unity quantum interference in compact…","keywords":["Hong-Ou-Mandel interference","transverse spatial modes","inverse design","integrated photonics","mode beamsplitter","silicon photonics","multimode waveguide","quantum interference"],"falsifier":"Measure the mode content of the device output with a mode-resolving demultiplexer or near-field imaging while injecting one photon per input mode; if a non-negligible fraction of the power appears in TE2, TE3, radiation modes, or as multiplexer cross-talk, the two-port model behind Eq. (5) would not describe the interference and the reported visibility would need reinterpretation.","tokens_in":10090,"feed_emoji":"⚛️","tokens_out":8806,"duration_ms":78044,"temperature":0.7,"pith_summary":"This paper reports that a silicon photonic component produced by inverse design—a beamsplitter that operates on the transverse spatial mode of light in a waveguide, not on separate paths—can create near-unity Hong-Ou-Mandel interference between two photons. The best device shows a measured visibility of 99.56±0.64 percent, three copies of the same design average 99.38±0.41 percent, and every tested design exceeds 98 percent, all in a device footprint of 3 µm by 3 µm. The claim matters because transverse spatial modes are a relatively unused degree of freedom in integrated quantum photonics, and this result suggests inverse-designed components can deliver the compact, reproducible mode mixing needed to exploit them.","feed_headline":"Mode beamsplitter hits 99.56% quantum visibility in 3 µm","feed_subtitle":"Every inverse-designed mode beamsplitter exceeds 98 percent visibility, and copies of the best design average 99.38 percent.","key_machinery":"The central object is an inverse-designed $3\\,\\mu\\mathrm{m} \\times 3\\,\\mu\\mathrm{m}$ mode beamsplitter that mixes the two lowest transverse electric modes (TE0 and TE1) of a multimode silicon waveguide. Its topology is produced by adjoint-method topology optimization that minimizes the squared deviation of simulated S-parameters from a target transmission of $\\sqrt{0.5}$ for each input–output pair, with level-set constraints on minimum feature size; the same optimizer also produces the mode multiplexers and demultiplexers. The theoretical analysis compresses the device into a two-port transfer matrix with transmission amplitudes $t_i$, reflection amplitudes $r_i$, and three phases, leading to an effective splitting ratio $\\eta_{\\rm eff}=t_0^2/(r_0^2+t_0^2)$ and a predicted Hong-Ou-Mandel visibility $V_{\\rm HOM}=-2\\eta_{\\rm eff}(1-\\eta_{\\rm eff})\\cos\\alpha\\,I_{\\rm overlap}(0)/(1-2\\eta_{\\rm eff}+2\\eta_{\\rm eff}^2)$, which is the formula against which the measurements are checked.","core_discovery":"At the paper's core is the demonstration that inverse-designed transverse-mode beamsplitters behave as high-quality Hong-Ou-Mandel interferometers. Using a device that couples the TE0 and TE1 modes of a multimode silicon waveguide, the authors measure V_HOM = 99.56 ± 0.64 percent for their best copy, with an average of 99.38 ± 0.41 percent over three copies of the same design; the three different inverse-designed beamsplitters all exceed 98 percent. Measured visibilities agree with the prediction of a lossy, unbalanced, asymmetric beamsplitter model, Eq. (5), based on the classically measured effective splitting ratio; the best device, at η_eff = 48.77 percent, would reach about 99.95 percent with a perfect source. The paper positions this as evidence that inverse-designed multimode components are suitable building blocks for compact integrated quantum photonic circuits.","pith_inferences":["By extension, the near-source-limited visibility implies that a straightforward test—repeating the experiment with a higher-indistinguishability source on the same chip—should raise measured values above 99.9 percent; the paper does not run that test.","By extension, the same 3 µm design area and optimization recipe could plausibly be scaled to more than two transverse modes for qudit operations, but the paper only demonstrates two-mode mixing and a simulated three-port tritter, so this remains an extrapolation.","By extension, if balanced loss is all that matters for HOM visibility, then quantum components may tolerate higher insertion loss than classical components in the same platform, which could relax fabrication constraints for multimode quantum devices."],"forward_implications":["Transverse spatial modes can be used as a practical degree of freedom for integrated quantum photonic circuits, adding Hilbert-space dimension without adding extra waveguide paths.","The same inverse-design procedure can produce different splitting ratios and mode counts in the same footprint; the paper reports a three-port beamsplitter designed in the same area.","Because the visibility depends only on the effective splitting ratio and the phase $\\alpha$, balanced losses do not destroy the quantum interference, so lossy but balanced devices remain useful.","Simulated fabrication robustness keeps theoretical visibility above 85 percent for a ±5 nm bias and above 90 percent for ±2.5 nm, indicating that near-unity interference survives realistic manufacturing tolerances.","The measured visibilities are close to the source limit, so improving the photon-pair source should push the on-chip visibility toward the predicted 99.95 percent without changing the device."],"supporting_citations":[{"why":"defines Hong-Ou-Mandel interference, the phenomenon the paper measures","marker":"[24]"},{"why":"previously demonstrated transverse-mode HOM interference with a Bragg-grating beamsplitter, the baseline for size and parity restrictions","marker":"[25]"},{"why":"supplies the inverse-design software and adjoint optimization approach used to generate the devices","marker":"[27]"},{"why":"provides the coincidence-probability model and fit function used to extract HOM visibility from the measured dips","marker":"[39]"},{"why":"supports the paper's claim that optimized beamsplitters acquire the phase differences required by energy conservation","marker":"[41]"},{"why":"provides the level-set fabrication constraints that make the optimized topologies manufacturable","marker":"[42]"}],"fun_headline_variants":["3-µm inverse-designed beamsplitter hits 99.56% quantum visibility","Smallest mode beamsplitter delivers 99.56% quantum interference","99.56% HOM visibility from a 3-µm on-chip mode splitter","Near-unity quantum interference from a tiny inverse-designed chip"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The results stand on the assumption that the device behaves as a lossy two-port beamsplitter that only mixes the two lowest waveguide modes, with the multiplexers delivering one clean photon in each mode; if appreciable power leaks into higher modes or radiation, or cross-talk enters at the multiplexers, the measured visibility no longer isolates two-mode interference.","fun_headline_variants_meta":{"raw":{"variants":["3-µm inverse-designed beamsplitter hits 99.56% quantum visibility","Smallest mode beamsplitter delivers 99.56% quantum interference","99.56% HOM visibility from a 3-µm on-chip mode splitter","Near-unity quantum interference from a tiny inverse-designed chip"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000642,"raw_usage":{"total_tokens":2941,"prompt_tokens":918,"completion_tokens":2023,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":534,"completion_tokens_details":{"reasoning_tokens":1942}},"tokens_in":534,"tokens_out":2023,"duration_ms":14841,"temperature":1.0,"reasoning_tokens":1942,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:48:55.856148+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the mode content of the device output with a mode-resolving demultiplexer or near-field imaging while injecting one photon per input mode; if a non-negligible fraction of the power appears in TE2, TE3, radiation modes, or as multiplexer cross-talk, the two-port model behind Eq. (5) would not describe the interference and the reported visibility would need reinterpretation.","supporting_citations":[{"cited_title":"Measurement of subpicosecond time intervals between two photons by interference,","cited_arxiv_id":null,"evidence_quote":"defines Hong-Ou-Mandel interference, the phenomenon the paper measures"},{"cited_title":"Quantum interference between transverse spatial waveguide modes,","cited_arxiv_id":null,"evidence_quote":"previously demonstrated transverse-mode HOM interference with a Bragg-grating beamsplitter, the baseline for size and parity restrictions"},{"cited_title":"Nanophotonic inverse design with spins: Software architecture and practical considerations,","cited_arxiv_id":null,"evidence_quote":"supplies the inverse-design software and adjoint optimization approach used to generate the devices"},{"cited_title":"Anti-symmetrization reveals hidden entanglement,","cited_arxiv_id":null,"evidence_quote":"provides the coincidence-probability model and fit function used to extract HOM visibility from the measured dips"},{"cited_title":"Exploring the fundamental limits of integrated beam splitters with arbitrary phase via topology optimization,","cited_arxiv_id":null,"evidence_quote":"supports the paper's claim that optimized beamsplitters acquire the phase differences required by energy conservation"},{"cited_title":"Analytical level set fabrication constraints for inverse design,","cited_arxiv_id":null,"evidence_quote":"provides the level-set fabrication constraints that make the optimized topologies manufacturable"}],"review_version":1}