REVIEW 3 major objections 5 minor 2 references
Fabrication of a 3D mode size converter for efficient edge coupling in photonic integrated circuits
T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read A silicon external mask can create a vertical thickness taper that, combined with a lithographic width taper, forms a 3D mode size converter coupling lensed fibers to LNOI waveguides at about 1 dB per facet.
desk verdict A useful fabrication variation for LNOI edge couplers, with reported losses that are plausible but carry an unquantified systematic uncertainty from the extraction method. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing mechanism is the external Si mask used during ICP-RIE etching: the mask sits roughly 1 mm above the LNOI film, so the plasma etching gas spreads under its edge and thins the film gradually, producing a smooth slope rather than a step. This local area etch creates two film thicknesses on one chip with no additional lithography; then ordinary electron-beam lithography defines a width-taper pattern, and a global etch produces rib-type waveguides in the thick region and strip-type waveguides in the thin region. The combination of the height taper and width taper forms the 3D mode size converter that adiabatically transforms the mode. Performance is extracted with the Fabry-Perot Airy distribution for a lossy medium (Eq. 2), using ring-resonator intrinsic Q values to estimate the propagation loss of the long waveguide.
What would settle it
Fabricate two waveguides with identical 3D couplers but different lengths, measure their Fabry-Perot transmission fringes, and solve for coupling loss and propagation loss simultaneously without using the ring-resonator loss estimate; if the extracted per-facet losses deviate from 1.16 dB (TE) and 0.71 dB (TM) by more than about 0.3 dB, the ring-resonator-only propagation-loss assumption is the source.
Extended reading notes
Core claim
The central claim is that a vertical thickness taper can be made on an LNOI chip by local-area dry etching with an external Si mask, and that this height taper combined with a lithographic width taper acts as a 3D mode size converter that couples light between a lensed fiber and a rib waveguide with low loss. The best measured coupling losses are approximately 1.16 dB/facet for TE and 0.71 dB/facet for TM at 1550 nm, close to numerical mode-overlap simulations. The local etch produces a smooth 425 nm height change over 180 µm (a slope of about 0.14°), from 600 nm down to 175 nm film thickness, while the width taper spans 100 µm. In the final device, region (i) is a 2-µm-wide rib waveguide in 600 nm film, region (ii) is the 3D converter, and region (iii) is a strip waveguide at the cleaved facet; the converter changes both height and width so that the guided mode grows to match the fiber mode.
Load-bearing premise
The reported coupling losses assume the tapered mode-converter sections add no propagation loss beyond the loss measured in the thick waveguide region; if the sloped or narrow-tip sections lose extra light, the per-facet dB values are systematically off.
Editorial extensions
If this is right
- If the reported values are correct, LNOI edge coupling can be brought to about 1 dB per facet with a fabrication flow that needs no CMP step and no extra lithography beyond the standard waveguide patterning.
- The match between measured coupling loss and mode-overlap simulation supports the claim that the low loss comes from the 3D mode conversion itself rather than from a fortunate cleave or alignment.
- Because local-area etching is performed before global patterning and uses a cleanroom-compatible Si mask, the process is compatible with wafer-scale fabrication; the authors report higher yield when local etching comes first.
- With accurate deep-RIE dicing and thicker bottom cladding, the authors expect the lossy inverse-taper region to shrink, which would allow flat fiber-array packaging and multi-channel operation.
- The same vertical-taper mechanism could apply to other thin-film photonic platforms, since the slope is defined geometrically by the mask gap and etch rather than by the specific material.
Reading between the lines
- The paper does not pursue a polarization-diverse design, but its own data show TE is best at roughly 500 nm taper width while TM is best near 3 µm; a single taper width can therefore favor one polarization, and a polarization-independent coupler would need separate tapers or a modified cross-section.
- The quoted dB/facet numbers inherit the uncertainty of using ring-resonator loss in region (i) as the loss of the whole 10.5 mm waveguide; the paper's reported up-to-0.3 dB discrepancy between max- and min-based estimates gives a rough bound on that uncertainty, and a direct cut-back test on the converter alone would sharpen it.
- Because the slope angle is set by mask gap, etch time, and pressure, the method could be tuned to other vertical taper profiles without changing lithography, which suggests a systematic parameter sweep across gap and pressure would reveal how robust the 0.14° slope is.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper reports a 3D mode size converter on the lithium-niobate-on-insulator platform, fabricated by first thinning selected areas of the LN film with a Si external mask and then patterning the waveguide and width taper by lithography and global etching. This produces a vertical thickness taper plus a lateral width taper without an additional lithography step. The authors characterize the resulting edge couplers by Fabry-Perot transmission measurements on a 10.5 mm waveguide with converters at both ends, using ring-resonator Q-factors for the propagation loss and Lumerical simulations for the effective indices and the mode-overlap coupling loss. They report approximately 1.16 dB/facet for TE and 0.71 dB/facet for TM at 1550 nm, with the best performance at taper widths of 500 nm and 3 um, respectively. The paper also gives profilometry of the vertical slope, SEM cross-sections, FSR comparison, and Q-factor data for the ring resonators.
Significance. If the reported numbers are reliable, the fabrication method is a useful addition to the LNOI toolbox: it replaces CMP or multilayer lithography with a single external-mask dry-etch step while still producing a 3D taper, and it has plausible wafer-scale potential. The claimed coupling losses are competitive with published LNOI edge couplers, and the paper supports the claim with several independent measurements (thickness profile, SEM, FSR, Q-factor) as well as numerical mode-overlap simulations. The main weakness is that the quantitative extraction of the per-facet coupling loss rests on a single ring-resonator propagation-loss value applied to the whole waveguide, including the tapered and strip regions that the paper itself identifies as lossy; this leaves a systematic uncertainty in the headline dB/facet values that is acknowledged but not quantified.
major comments (3)
- [Section 5, Eqs. (2) and (3)] The reported per-facet coupling loss is directly sensitive to the total propagation loss alpha*l used in Eq. (2), but alpha is taken exclusively from a ring resonator in region (i) with a 2-um-wide, 600-nm-thick rib and 200-nm slab. Regions (ii) and (iii) are excluded from the loss estimate, and the Discussion explicitly calls the inverse-taper region lossy. The paper itself acknowledges up to 0.3 dB discrepancy between the max- and min-based estimates and attributes it to the difference between ring-derived loss and the actual waveguide loss including the mode size converter. Since Eq. (3) shifts the extracted coupling loss by half of any error in A_T, this unquantified component is a large fraction of the 1.16 dB/facet (TE) and 0.71 dB/facet (TM) claims. Please provide a quantitative bound on the loss contribution of regions (ii) and (iii), for example by cutback measurements, by measuring waveguides with different converter lengths, or by a conservative upper-bound estimate, and restate the headline values with that systematic uncertainty.
- [Section 5, Fig. 4(f)] The headline numbers are selected as the best values from a taper-width scan, but the manuscript does not state how many devices or chips were measured, how many repeated measurements were made, or what the device-to-device variation is. Figure 4(f) shows the max- and min-based estimates without error bars or confidence intervals. Because the claim is a quantitative performance figure, the authors should report the number of measured devices, the distribution of extracted coupling losses, and the resulting uncertainty on the reported dB/facet values.
- [Section 5, Eqs. (1) and (5), and Section 4] The Fabry-Perot extraction uses the simulated effective index of region (iii) to compute the facet reflectance R in Eq. (1). The agreement claimed between measured and simulated coupling loss is therefore not fully independent of the simulation used for the mode-overlap calculation, and the sensitivity of the extracted coupling loss to the assumed n_eff is not reported. Please provide the assumed n_eff values and an error-propagation estimate for R. In addition, the length of region (i) is given as approximately 1.05 cm in Section 4 but as 9.5 mm in Section 5; since the propagation term in Eq. (2) depends linearly on the total length, this inconsistency should be resolved and the exact value of l used in the extraction stated.
minor comments (5)
- [Section 4] The waveguide length description is inconsistent: region (i) is stated as 1.05 cm, while Section 5 says the lengths of regions (i) and (iii) are 9.5 mm and 1 mm; please clarify which length corresponds to the fabricated device and to the value of l used in Eq. (2).
- [Section 5, Eq. (2)] The notation A_T,max(min) and T_max(min) should be defined more explicitly, and the equation should state the units of each term; currently A_T is labeled as dB but T_max(min) appears to be in dB as well, and the split in Eq. (3) relies on this convention.
- [Abstract and Section 1] The phrase 'edge coupling efficiency' is used where the reported quantity is a coupling loss in dB/facet; please use consistent terminology to avoid sign ambiguity.
- [Section 2] The statement that the process is 'without additional lithography' should be clarified, since electron-beam lithography with HSQ is still used for waveguide patterning; the novelty is that the vertical thickness variation requires no additional lithography step beyond the standard patterning.
- [References] Reference [41] appears incompletely formatted ('Quantum Electron. 72 I: Passive-resonator linewidth...'); please correct the bibliographic entry.
Circularity Check
No significant circularity: coupling-loss extraction uses measured fringes with independently estimated propagation loss and simulated refractive index, none fitted to the claimed result.
full rationale
The paper's central result—approximately 1.16 dB/facet TE and 0.71 dB/facet TM coupling loss—is obtained by inserting measured transmission maxima/minima, ring-resonator-derived propagation loss, and simulated effective refractive indices into the Fabry-Perot equation (Eq. 2). The simulated n_eff is used to compute facet reflectance and FSR, but it is an auxiliary input, not a fitted parameter, and the measured coupling loss is not derived from the simulated coupling loss in Fig. 4(e); the agreement between them is a post hoc comparison. The propagation loss is taken from region (i) only, and the paper explicitly acknowledges discrepancies up to 0.3 dB between max- and min-based estimates due to differences between ring-derived loss and actual waveguide loss including the mode size converter. This is a stated approximation and potential systematic error in accuracy, not circularity, because it does not make the conclusion equivalent to its input by construction. No load-bearing self-citation chain or imported uniqueness theorem appears; the cited prior work by overlapping authors is background context only. The derivation is therefore self-contained with respect to circularity, although the reported dB/facet values carry an acknowledged unquantified systematic uncertainty from the propagation-loss approximation.
Assumptions & free parameters
free parameters (3)
- Taper width at the chip edge =
500 nm for best TE; 3 um for best TM
- Effective refractive indices n_eff for facet reflectance =
Simulated values in Fig. 4(e), e.g., 1.92 and 1.68 for TE/TM in region (iii)
- Propagation loss alpha =
0.75 dB/cm (TE), 1.67 dB/cm (TM) from ring resonator intrinsic Q
assumptions (4)
- domain assumption Propagation loss measured from a ring resonator in region (i) applies to the entire waveguide, including the sloped taper and strip region.
- domain assumption The Fabry-Perot interference model with constant facet reflectance R (from simulated n_eff) and constant alpha describes the measured transmission fringes.
- domain assumption The 0.14 degree slope and width taper give an adiabatic mode conversion with negligible transition loss.
- domain assumption Lumerical mode simulation accurately predicts effective indices and coupling losses for the fabricated cross-sections.
Cite this review
Pith. "Pith review of Fabrication of a 3D mode size converter for efficient edge coupling in photonic integrated circuits." pith.science (2026). https://pith.science/paper/WI5A4OE2
@misc{pith2026241116221,
author = {Pith},
title = {Pith review of: Fabrication of a 3D mode size converter for efficient edge coupling in photonic integrated circuits},
year = {2026},
howpublished = {\url{https://pith.science/paper/WI5A4OE2}},
note = {Machine review of arXiv:2411.16221}
}
read the original abstract
We demonstrate efficient edge couplers by fabricating a 3D mode size converter on a lithium niobate-on-insulator photonic platform. The 3D mode size converter is fabricated using an etching process that employs a Si external mask to provide height variation and adjust the width variation through tapering patterns via lithography. The measured edge coupling efficiency with a 3D mode size converter was approximately 1.16 dB/facet for the TE mode and approximately 0.71 dB/facet for the TM mode at a wavelength of 1550 nm.
Figures
Reference graph
Works this paper leans on
-
[1]
Lithium niobate photonics: Unlocking the electromagnetic spectrum,
1 A. Boes, L. Chang, C. Langrock, et al., “Lithium niobate photonics: Unlocking the electromagnetic spectrum,” Science 379(6627), eabj4396 (2023). 2 A. Boes, B. Corcoran, L. Chang, et al., “Status and potential of lithium niobate on insulator (LNOI) for photonic integrated circuits,” Laser Photonics Rev. 12(4), 1700256 (2018). 3 C. Wang, M. Zhang, X. Chen...
work page 2023
-
[2022]
High-efficient coupler for thin-film lithium niobate waveguide devices,
1(1), 016001–016001. 34 C. Hu, A. Pan, T. Li, et al., “High-efficient coupler for thin-film lithium niobate waveguide devices,” Opt. Express 29(4), 5397–5406 (2021). 35 P. Ying, H. Tan, J. Zhang, et al., “Low-loss edge-coupling thin-film lithium niobate modulator with an efficient phase shifter,” Opt. Lett. 46(6), 1478–1481 (2021). 36 D. Jia, Q. Luo, C. Y...
work page 2021
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.