{"id":"2941bdbb-2189-4831-88d4-e92297c62a82","arxiv_id":"2507.01874","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Graded silicon beam lattices with tens of thousands of unit cells are inversely designed with ray tracing and fabricated by wafer-scale photolithography, guiding MHz elastic waves along a designed figure-eight path.","lead":"This paper combines a ray-tracing design tool with silicon chip fabrication to build a phononic metamaterial waveguide containing tens of thousands of tiny mechanical units. The authors demonstrate that elastic waves can follow a designed figure-eight path on a wafer, and argue the same process can scale to hundreds of thousands or millions of unit cells.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'hundreds of thousands of unit cells' realization claim rests on companion [32] and a ray-only 143k design, not on data in this preprint; conditional verdict is appropriate until that evidence is verified.","rationale":"The reader's weakest-assumption analysis focuses on ray-tracing smoothness. That is a real premise, but the manuscript provides substantial evidence for it in the demonstrated regime: transient FE simulations and experimental line scans (Fig. 4; SI Sec. 5) agree well for the 192×192 figure-eight across 250-800 kHz, including multiple tile-boundary crossings. I therefore do not see the smoothness concern as the decisive issue. The more load-bearing gap is evidentiary: the abstract promises realization at hundreds of thousands of cells, while the only fully validated realization in this paper is ~37,000 cells. The 143,000-cell example is ray-tracing only, and the 600,000-cell wafer is cited to a companion paper. The conditional verdict is appropriate: if [32] and the repository provide the missing guided-wave data, the central claim is supported; if not, the claim should be narrowed. This is a reporting/reproducibility concern rather than a demonstrated physical flaw, so it does not warrant rejection, only a condition on the missing evidence.","tokens_in":19398,"tokens_out":8593,"duration_ms":111083,"concrete_test":"Obtain companion manuscript [32] or the dataset at doi:10.3929/ethz-b-000742467 and verify that it contains an experimental guided-wave measurement for the 600,000-cell wafer, including line-scan or field data comparable to Fig. 4; if no such measurement exists, the 'hundreds of thousands' realization claim should be removed from the abstract and the paper's central claim should be limited to the tens-of-thousands scale demonstrated in this manuscript.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim is that the framework enables design and realization of waveguides with hundreds of thousands of unit cells. In this manuscript, the only fabricated and experimentally validated waveguide is the 192×192 (~37,000-cell) figure-eight (Fig. 2c; Fig. 4). The 256×256 cross is validated only by simulation, and the ~143,000-cell assembly in SI Fig. S2 is presented as θ distributions and ray paths without a transient FE wavefield. The 600,000-cell wafer cited in the Microfabrication section is attributed to companion manuscript [32], so the 'realization' part of the headline claim cannot be checked from this preprint or its linked repository alone. A secondary, related technical risk is that tile scalability assumes θ-continuity at tile boundaries; the ray equations (S1)-(S2) involve ∇θ through ∂ω/∂x, and continuity of θ alone does not guarantee smoothness of ∇θ. The demonstrated 3×3 assembly agrees with FE and experiment, which mitigates this for that design, but it does not establish that a much larger assembly with many boundary crossings behaves as ray-traced.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents a framework for inverse design of graded phononic metamaterials based on ray tracing and modular tile assembly, combined with a silicon-on-insulator wafer microfabrication process to create free-standing architected films. The authors design two tile types by optimizing ray trajectories, assemble them into figure-eight (192×192 unit cells) and cross (256×256 unit cells) patterns, and validate the figure-eight design experimentally using a pump-probe laser/interferometer setup. Experimental line-scan and frequency-resolved data show good agreement with finite-element simulations over a broad band, and the authors claim that the design/fabrication framework enables waveguides with hundreds of thousands of unit cells and the potential to scale to millions.","tokens_in":19528,"tokens_out":13393,"duration_ms":141404,"significance":"If its claims hold, this work is a significant advance because it combines an efficient ray tracing inverse-design method with a wafer-scale microfabrication route for free-standing elastic metamaterials, a combination that is rare and of broad interest for MEMS-scale wave manipulation. The experimental validation is a genuine strength: the agreement between the measured line scans (Fig. 4) and the independent finite-element simulations, including the frequency-resolved comparisons in SI Sec. 5, provides strong evidence that the fabricated 36,864-cell waveguide behaves as designed, with no apparent parameter fitting in the experimental channel. The public code/data repository supports reproducibility. The central caveats are that the headline scale claim (hundreds of thousands of realized unit cells) is not supported by data in this manuscript, and the ray tracing smoothness assumption is not established for large tile assemblies.","major_comments":[{"comment":"The abstract states that the framework 'enables the design and realization of complex waveguides including hundreds of thousands of unit cells,' but the experimental realization in this manuscript is a 192×192 (36,864-cell) figure-eight (Fig. 2c, Fig. 4). The 256×256 cross (~66,000 cells) is validated only by finite-element simulation (SI Fig. S4), and the ~143,000-cell assembly in SI Fig. S2 is presented only as a θ distribution and ray paths, with no transient wavefield simulation. The ~600,000-cell wafer mentioned in the Microfabrication section is attributed to companion reference [32], which is not part of this submission and is not available for assessment. The 'realization' half of the headline claim is therefore not supported by evidence in this manuscript; the text should clearly separate what is demonstrated here (a ~37,000-cell prototype with experimental validation) from what is claimed via the companion paper or via ray-only designs.","section":"Abstract; Microfabrication (main text, p. 8); SI Sec. 1.4"},{"comment":"The ray tracing model assumes 'smooth spatial gradings and hence locally an approximately periodic medium' (Introduction), and the ray equations (S1)-(S2) involve ∂ω/∂x = (∂ω/∂θ)(∂θ/∂x). The tile assembly enforces continuity of θ at tile boundaries (θ = 0.4L on each tile perimeter) but does not enforce continuity of ∇θ, so the normal derivative of θ can jump at tile boundaries. Such a jump produces a discontinuity in k̇, which is inconsistent with the smooth-grading assumption on which ray tracing is based. The authors do not quantify the gradient jumps at tile boundaries in the 3×3 and 4×4 assemblies, nor do they validate the ~143,000-cell assembly (SI Fig. S2) with a wavefield simulation. The good FE/experiment agreement for the 3×3 figure-eight mitigates the concern for that particular design, but it does not establish that much larger assemblies with many tile boundaries are accurately predicted by ray tracing. The authors should either enforce or verify C1 continuity of θ across tile boundaries, or validate a large assembly with transient FE.","section":"Introduction (p. 3); SI Sec. 1.2 (Eqs. S1-S2); SI Sec. 1.4"},{"comment":"The description of the design-variable interpolation is incomplete. The text states that the design grid is refined from a spacing of 16L to L and that unit cell designs are 'interpolated' from the design grid, but it does not specify the interpolation scheme (e.g., piecewise linear, spline) or how the spatial derivative ∂θ/∂x is evaluated in the ray tracing system (S1)-(S2) and in the adjoint gradient (S11). If the final design is piecewise constant per unit cell, then ∂θ/∂x is a sum of delta functions and the ray tracing equations are not well-defined; if a smooth interpolation is used, the manuscript should state it. This detail is essential for reproducing the optimization and for assessing the smoothness assumption raised in the preceding comment.","section":"SI Sec. 1.3"}],"minor_comments":[{"comment":"The text says the structures span 'three orders of magnitude in length scales'; the fabricated figure-eight spans 3 cm with L = 100 µm (ratio 300), while the 100-mm wafer with 100-µm cells (ratio 1000) comes from companion [32]. Please specify which structure is meant and avoid overstating the demonstrated range.","section":"Introduction, Conclusion"},{"comment":"Reference [32] is listed as a companion manuscript without a preprint identifier or status note; since the 600,000-cell claim rests on it, readers need a link or a clear statement of its availability.","section":"References"},{"comment":"The manuscript does not report the values of w1 and w2 used in Eq. (S3) for the two tile designs; please provide them for reproducibility.","section":"SI Sec. 1.3"},{"comment":"The text says two measurement points have poor signal-to-noise ratio 'marked by the black arrow for line scan L2,' but the figure appears to show only one arrow; please clarify the notation.","section":"Fig. 4b"},{"comment":"The transient FE simulation uses a 1 µs half-sine excitation and states that it 'spans approximately the same frequency range' as the experimental 1 ns laser pulse. Please quantify the spectral content of the experimental acoustic pulse or justify why the approximation is adequate for the comparisons in Fig. 4 and SI Sec. 5.","section":"SI Sec. 2.2"}],"recommendation":"major_revision","confidential_remarks":"The overclaim in the abstract and conclusion is the main concern: the 'design and realization of hundreds of thousands of unit cells' is not demonstrated in this manuscript. Please ask the authors to either temper the claim or include sufficient evidence (for example, a summary of the companion [32] 600k-cell characterization, or a transient FE simulation of the 143k design). The smoothness concern about tile boundaries is real; a quantitative study of gradient continuity at tile boundaries would materially strengthen the scalability statement."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe real content here is the experimental demonstration: a free-standing silicon phononic film with a 192x192-cell graded figure-eight pattern, fabricated by photolithography and DRIE on an SOI wafer, and probed with a pulsed-laser pump and heterodyne interferometer. The line-scan data match finite element simulations well over a broad band, and the wave guiding is convincing. That part holds up.\n\nWhat's new is mostly on the fabrication side. Making free-standing architected films with tens of thousands of unit cells via wafer-scale processes is a genuine practical advance over the usual 3D-printed or two-photon lattices. The modular tile assembly is a useful extension of the authors' earlier ray-tracing work, and the k-dependent cost function and arbitrary exit curve are sensible increments. The experimental agreement also breaks the mild circularity of using the same FE dispersion model for both design and validation.\n\nWhere I agree with the conditional verdict: the abstract's \"hundreds of thousands of unit cells\" claim is not demonstrated in this preprint. The fabricated prototype is about 37,000 cells. The 256x256 cross is FE-only, and the ~143,000-cell assembly in SI Fig. S2 shows ray paths but no transient wavefield. The 600,000-cell wafer is attributed to companion manuscript [32], so that part of the headline claim can't be checked from this paper alone. This needs to be fixed, either by including the supporting evidence or by softening the claim.\n\nA more technical soft spot: tile assembly guarantees continuity of the design parameter theta across tile boundaries, but the ray equations depend on the gradient of theta through d(omega)/dx. Continuity of theta alone doesn't ensure smoothness of that gradient. The 3x3 assembly agrees well with FE and experiment, which mitigates this for that design, but larger assemblies with many boundary crossings could behave differently. A comment or a numerical check would help.\n\nMinor issues: the cost weights w1 and w2 aren't reported, which makes reproduction harder, but that's not a deep flaw.\n\nBottom line: a solid applied-physics paper with an honest experimental core and an overstated headline generality. It deserves a serious referee, and I'd recommend pushing for revision that either validates or scales back the large-scale claim.\n\nRecommendation: send to peer review.","headline":"A solid experimental demonstration of wave guiding in a microfabricated silicon phononic film, but the headline 'hundreds of thousands of unit cells' claim is not supported by the data in this preprint.","tokens_in":20150,"tokens_out":2138,"would_cite":true,"duration_ms":23701,"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":"Graded phononic waveguides with hundreds of thousands of unit cells can now be designed and fabricated on a silicon wafer.","keywords":["phononic metamaterials","spatially graded","ray tracing","inverse design","microfabrication","silicon-on-insulator","wave guiding","broadband"],"falsifier":"Fabricate a graded silicon film with a sharp gradient across just a few unit cells, launch the same broadband laser pulse, and compare the measured displacement along a designed ray path with finite element simulation; a large deviation or loss of guiding would indicate where the local-periodicity assumption breaks down.","tokens_in":19111,"feed_emoji":"🔬","tokens_out":9695,"duration_ms":82866,"temperature":0.7,"pith_summary":"The paper aims to remove the scalability bottleneck that has kept phononic metamaterials at toy scales: it pairs an inverse design method that handles hundreds of thousands of unit cells with a silicon-wafer fabrication route that can realize them as free-standing films. The design side uses ray tracing through locally periodic Bloch dispersion relations, so optimization cost scales with the number of rays rather than unit cells, and a modular tile library lets large waveguides be assembled from small designed building blocks. The fabrication side adapts photolithography and deep reactive ion etching of silicon-on-insulator wafers to produce free-standing architected films, demonstrated here with roughly 600,000 unit cells on a 100 mm wafer. The central experimental claim is that a designed figure-eight waveguide, targeted at 750 kHz, actually guides broadband elastic waves along the intended path in the fabricated sample, with measured displacement fields matching finite element simulations.","feed_headline":"One ray-tracing pipeline scales phononic waveguides to wafers","feed_subtitle":"Inverse design plus silicon etching yields free-standing films with ~600,000 unit cells that guide broadband elastic waves on chip.","key_machinery":"Ray tracing in graded metamaterials, governed by the Hamiltonian system $\\dot{x} = \\partial \\omega/\\partial k$ and $\\dot{k} = -\\partial \\omega/\\partial x$, uses the local dispersion relation of the beam unit cell as a Hamiltonian to compute trajectories; an adjoint-state optimization shapes these rays to prescribed exit positions and wave vectors. The modular tile library then assembles individually designed tiles, with matching boundary values of $\\theta$, into larger waveguides without recomputation.","core_discovery":"The paper demonstrates an end-to-end path from computer-aided design to physical sample for spatially graded phononic metamaterials at a scale that was previously impractical. Its central discovery is that a ray tracing model, built on locally computed Bloch dispersion surfaces for a beam-lattice unit cell parameterized by one angle $\\theta$, is accurate and fast enough to serve as the forward model for inverse design of waveguides spanning tens of thousands of unit cells, and that the resulting designs can be fabricated by standard semiconductor processing into free-standing silicon films. Two tile types are designed—one converting a point excitation into four outgoing plane waves, the other rotating an incident plane wave by $90^\\circ$—and assembled into figure-eight and cross waveguides. Transient finite element simulations reproduce the ray-predicted paths, and laser-excited interferometric measurements on the fabricated figure-eight wafer confirm wave guiding over a band from about 250 to 800 kHz, well beyond the 750 kHz design frequency.","pith_inferences":["The same design pipeline could optimize higher dispersion branches and out-of-plane or in-plane targets, suggesting extensions to multi-mode or topology-switching waveguides.","The ray tracing forward model could be tested more aggressively at sharper-than-demonstrated gradings and at higher frequencies, where corners and amplitude effects are more pronounced.","Because the ray-optimized paths are computed at a single frequency, one could test whether multi-frequency or broadband-weighted cost functions produce even wider operational bands.","The combination of a general ray model with a standardized tile library points toward reusable 'metamaterial standard cells' analogous to logic gates in electronics, though the paper does not pursue this analogy."],"forward_implications":["Waveguide designs spanning hundreds of thousands to millions of unit cells become computationally tractable, since the optimization cost scales with the number of rays rather than the number of cells.","Semiconductor wafer processing turns metamaterial fabrication into a batch, chip-style process, enabling many samples per wafer and direct integration with MEMS and on-chip signal-processing devices.","The demonstrated broadband guiding suggests that target-frequency designs may inherit a wide operational band, easing practical deployment.","Free-standing architected silicon films open a route to studying wave attenuation in the architecture alone, in the absence of intrinsic material damping and substrate losses.","The modular tile approach allows new functions to be added by designing new tiles, so functionality can grow incrementally without redesigning the whole device."],"supporting_citations":[{"why":"Establishes the ray theory for elastic wave propagation in graded metamaterials used as the forward model.","marker":"[28]"},{"why":"Provides the prior inverse design formulation via ray tracing and the adjoint method that the tile optimization extends.","marker":"[29]"},{"why":"Supplies the beam finite element model for dispersion computations of the unit cell.","marker":"[30]"},{"why":"Validates the beam-element dispersion framework against experiments, supporting the design model.","marker":"[31]"},{"why":"Reports the wafer-scale fabrication of microarchitected films with hundreds of thousands of unit cells.","marker":"[32]"}],"fun_headline_variants":["Ray tracing and silicon etching scale phononic waveguides","Scalable inverse design plus microfabrication for phononics","Hundreds of thousands of cells: phononic waveguides on a wafer","Ray-tracing designs silicon phononic waveguides at scale","Wafer-scale phononic metamaterials via ray tracing and etching"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The ray tracing model is based on local Bloch analysis, so it assumes the spatial grading is smooth enough that neighboring unit cells are almost periodic; if the grading is too sharp, the predicted rays will not match the actual wave field.","fun_headline_variants_meta":{"raw":{"variants":["Ray tracing and silicon etching scale phononic waveguides","Scalable inverse design plus microfabrication for phononics","Hundreds of thousands of cells: phononic waveguides on a wafer","Ray-tracing designs silicon phononic waveguides at scale","Wafer-scale phononic metamaterials via ray tracing and etching"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000163,"raw_usage":{"total_tokens":1253,"prompt_tokens":968,"completion_tokens":285,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":200}},"tokens_in":584,"tokens_out":285,"duration_ms":3996,"temperature":1.0,"reasoning_tokens":200,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T20:42:04.043788+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fabricate a graded silicon film with a sharp gradient across just a few unit cells, launch the same broadband laser pulse, and compare the measured displacement along a designed ray path with finite element simulation; a large deviation or loss of guiding would indicate where the local-periodicity assumption breaks down.","supporting_citations":[{"cited_title":"& author Kochmann, D","cited_arxiv_id":null,"evidence_quote":"Establishes the ray theory for elastic wave propagation in graded metamaterials used as the forward model."},{"cited_title":"& author Kochmann, D","cited_arxiv_id":null,"evidence_quote":"Provides the prior inverse design formulation via ray tracing and the adjoint method that the tile optimization extends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the beam finite element model for dispersion computations of the unit cell."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Validates the beam-element dispersion framework against experiments, supporting the design model."},{"cited_title":", author Dorn, C","cited_arxiv_id":null,"evidence_quote":"Reports the wafer-scale fabrication of microarchitected films with hundreds of thousands of unit cells."}],"review_version":1}