Fabry-Perot-Insensitive Edge Coupling for Robust PIC Characterization

Alireza Geravand; David Turgeon; Raphael Dube-Demers; Sammy Noel Parise; Sophie LaRochelle; Wei Shi

REVIEW 5 major objections 6 minor 7 references

Fabry-Perot-Insensitive Edge Coupling for Robust PIC Characterization

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A quarter-wavelength shift between two edge-coupled measurements cancels the dominant Fabry-Perot ripple, reducing insertion-loss variability from 0.34 dB to 0.04 dB.

desk verdict A simple, parameter-free quarter-wave averaging trick that plausibly cancels fiber-chip Fabry-Perot ripple in edge-coupling measurements; the reported 0.34 to 0.04 dB reduction is promising but the experimental evidence is not yet airtight. read the letter →

arxiv 2506.08155 v1 pith:B7XIZBDD submitted 2025-06-09 physics.optics

Sammy Noel Parise , Raphael Dube-Demers , David Turgeon , Alireza Geravand , Sophie LaRochelle , Wei Shi This is my paper

classification physics.optics

keywords Fabry-Perotcavityedgecouplingphotonicintegratedcircuitsinsertionlosswafer-leveltestingcomplementarydualmeasurementalignmentrepeatability

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that the dominant variability in edge-coupled photonic chip measurements comes from a Fabry-Perot cavity formed between the fiber array and the chip surface, and that this variability can be largely removed by a complementary dual-measurement strategy. The authors show that subwavelength changes in the fiber-chip gap produce up to 0.34 dB of insertion-loss ripple. Shifting the gap by a quarter wavelength flips the sign of the leading interference term, so averaging the two spectra cancels it. Experimentally, the averaged spectra hold their spread to about 0.04 dB across a 1505-1675 nm sweep, and the same pairing improves alignment repeatability for coupling power and lateral position. If this holds, high-throughput wafer-level testing can avoid index-matching gel while keeping measurement precision.

What carries the argument

The machinery is the Taylor expansion of the single-cavity transmission T = T1 T2 / (1 + R1 R2 - 2 sqrt(R1 R2) cos delta) in powers of cos delta, together with the phase relation delta = 4 pi n d / lambda. Because the leading correction is proportional to cos delta, translating the fiber array by d' = d + lambda_0 / (4 n) makes that term change sign; averaging the two spectra then cancels it. The residual error is governed by the mismatch between the design wavelength lambda_0 and the actual wavelength during the sweep.

What would settle it

Measure the transmitted power through a fixed loopback while stepping the fiber-chip gap in sub-50 nm increments across at least one full wavelength, both with and without the complementary paired measurement. If the averaged spectrum still shows ripple correlated with the gap, or if the single-sweep ripple amplitude changes with absolute gap, then the assumption that translation only changes the phase is violated.

Watch

Extended reading notes

Core claim

The central claim is that the first-order Fabry-Perot modulation in edge-coupled insertion-loss measurements can be cancelled by averaging two measurements whose cavity lengths differ by a quarter wavelength. Starting from the single-cavity transmission formula, the authors expand in powers of cos delta; the leading term is a sinusoidal ripple of order 0.29 dB, and the second-order term contributes only about 0.018 dB. A quarter-wave translation changes the phase delta by pi, so cos delta changes sign, and the average removes the dominant ripple while leaving the second-order term nearly intact. The paper supports this with tunable-laser sweeps over 170 nm, reporting that the maximal spectral variation drops from 0.34 dB on single measurements to 0.04 dB after pairing, and that the Nelder-Mead alignment routine becomes more repeatable in coupling power and in the X coordinate.

Load-bearing premise

The cancellation assumes that translating the fiber array by a quarter wavelength changes only the Fabry-Perot phase, while the interface transmission coefficients and the lateral mode overlap between fiber and waveguide stay exactly the same.

Editorial extensions

If this is right

Wafer-level edge-coupled testing can be made repeatable without applying index-matching gel, removing a major practical bottleneck in high-volume photonic characterization.
The same quarter-wave pairing can be applied on existing test platforms with no additional hardware, since it only requires a precise relative translation of the fiber array.
Alignment routines such as Nelder-Mead become more reliable in coupling power and lateral position when fed with complementary paired measurements instead of single sweeps.
The residual error across a broad wavelength sweep remains near 0.04-0.05 dB when the design wavelength is chosen near the center of the band, which is sufficient for many precision photonics applications.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural extension would be to track the quarter-wave offset dynamically during a wavelength sweep, compensating for the residual error that currently arises when lambda differs from lambda_0.
The observed asymmetry in Y-axis alignment improvement suggests that angular misalignment, not just gap distance, also distorts the coupling landscape; a method that measures or corrects tip and tilt could extend the benefit to both axes.
The technique could be combined with loopback reference subtraction to separate the intrinsic device response from the residual cavity effect, potentially pushing the uncertainty below 0.04 dB.
Because the cancellation relies only on relative stage movement, it is compatible with automated probers that already have nanometer-resolution positioning, making it a low-friction add-on for industrial testing flows.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Here's my take. The core idea is genuinely new and simple: take two edge-coupling measurements with the fiber-chip gap shifted by a quarter wavelength, average them, and the first-order Fabry-Perot ripple cancels. The theory is clean and parameter-free, based on the standard FP transmission formula with tabulated Fresnel reflectivity. It predicts a residual error of about 0.04-0.05 dB over the 1505-1675 nm sweep, and the measured reduction from 0.34 dB to 0.04 dB matches that. No curve fitting, no circularity. That is a real contribution.

The practical value is also real: no new hardware, just a controlled relative displacement, and it addresses a genuine metrology problem in wafer-level PIC testing. The presentation is straightforward, and the paper does not oversell—it explicitly flags the Y-axis asymmetry as a hypothesis and mentions a control analysis.

The soft spots are in the experimental validation. The control analysis is described but never reported: the paper says each single measurement was averaged with the subsequent one to confirm the improvement isn't just from averaging, but no result, plot, or statistic is given. That is a genuine omission, because the 0.34-to-0.04 reduction is the headline claim. The preprocessing is also post hoc—outliers removed and data truncated to the first 700 points—which could bias the comparison. And the model assumes a 400 nm translation changes only the FP phase, not the mode overlap or effective reflectances. In a real edge coupler that may not hold exactly; the paper doesn't bound that residual. None of this sinks the central physics, but it means the claim of universal applicability is not yet established.

Who gets value: anyone working on automated edge-coupled PIC characterization, especially in industrial or wafer-level settings, will want to know about this. The paper deserves a serious referee. I'd send it out, with a request for the control data, raw or minimally processed spectra, and a quantitative estimate of the gap-dependent coupling error. If those come back clean, this is a solid, citable metrology result.

Referee Report

5 major / 6 minor

Summary. The paper proposes a complementary dual-measurement (CDM) technique to suppress Fabry-Perot (FP) interference ripples in edge-coupled photonic integrated circuit testing. By taking two transmission measurements with the fiber-chip gap shifted by λ/4, the first-order FP modulation reverses sign and cancels upon averaging. The authors derive a plane-wave FP model predicting a residual error of ~0.04–0.05 dB over a 1505–1675 nm sweep, and report an experimental reduction of insertion-loss variability from 0.34 dB to 0.04 dB. They also apply the technique to alignment optimization, claiming improved repeatability in coupling power and X-position (but not Y-position).

Significance. The technique is simple, requires no additional hardware, and is based on a transparent first-order cancellation argument. The theoretical residual estimate is parameter-free (using tabulated Fresnel coefficients) and is clearly derived. If the experimental validation holds up, the method would be directly useful for improving repeatability in wafer-level edge-coupled PIC characterization. However, the causal attribution of the observed reduction to FP cancellation is weakened by the unreported control analysis and the untested assumption of gap-independent coupling; these need to be addressed.

major comments (5)

[Section 3, Eq. (10)] The residual error expression is missing the factor 1/2 from the averaging in Eq. (9). With T and T' as in Eqs. (7)–(8), the average is T_mean = 0.93[1 + 0.032(cos δ + cos δ′) + ...], so the first-order residual coefficient should be 0.032, not 0.064. The subsequent estimates of 0.05 dB and 0.04 dB for λ = 1.505 µm and 1.675 µm are therefore a factor of 2 too large. This does not change the cancellation mechanism, but the quantitative prediction should be corrected.
[Section 5, control analysis paragraph] The control analysis described in Section 5 is not reported quantitatively; no figure, statistic, or p-value is given. Without this control, the claim that the error reduction in the alignment study is specifically due to FP cancellation rather than averaging cannot be evaluated. Please provide the control distributions and compare their standard deviations to the CDM and single-measurement cases.
[Section 3, Eqs. (1)–(9)] The derivation assumes that a translation of λ/4 changes only the FP phase δ, while the interface transmissions T1, T2 and the mode-overlap coupling remain unchanged. For a real edge coupler, the coupling coefficient can depend on the fiber-chip gap; a 400 nm shift may change the overlap integral and the effective reflectance. The paper does not bound the residual from these gap-dependent terms. I recommend measuring transmission as a function of gap over a range spanning several micrometers, separating the slow envelope from the FP ripple, and showing that the envelope variation over 400 nm is negligible (< 0.01 dB). Without this, the claimed universality of the technique for wafer-level probing is not established.
[Section 5, Y-axis asymmetry] The paper reports that the CDM technique did not improve repeatability along the Y axis and attributes this to a hypothesized angular misalignment. Because the conclusion claims that the technique improves alignment in the x-y plane, the unexplained Y-axis failure is a load-bearing limitation. Please provide a quantitative test of the hypothesis (e.g., measuring the angular alignment or performing the experiment after deliberate tilt correction), or at least state the conditions under which the technique is expected to work.
[Section 5, data preprocessing] The alignment analysis removes outliers outside ±2σ and truncates to the first 700 points. These choices can artificially reduce the reported standard deviations; the number of removed points and the sensitivity of the results to the truncation and outlier threshold are not reported. Please provide the raw data counts, the number of outliers removed, and a robustness check (e.g., varying the σ threshold and the truncation limit).

minor comments (6)

[Throughout] Typos: 'positionning' (Abstract and Introduction) should be 'positioning'; 'a serie of' should be 'a series'; 'FPs cavity' should be 'FP cavity'; 'substracting' should be 'subtracting'; 'a the CDM technique' should be 'a CDM technique'.
[Section 3, Eq. (11)] There is a typo in δ′(λ): it is written as (4π/λ)(d + λ0/λ) but should be (4π/λ)(d + λ0/4). As written, the equation is dimensionally inconsistent.
[Figure 3 caption] The caption for Fig. 3c ('Fig 3c shows a comparison in the fluctuation between two measurement at slightly difference distance using single and CDM measurement') is unclear; please rephrase to specify exactly what is plotted, e.g., the difference between two spectra for the single and CDM cases.
[Section 5, Fig. 3b] The text states 'Fig. 3b displays the CDM' but the figure shows the averaged spectra; please clarify in the caption that the CDM is the average of two quarter-wave-shifted measurements.
[References] Reference [2] lists 'Koen Alexander and et al.'; please provide the full author list or use a standard citation format.
[Section 5, signal processing] The width of the Gaussian filter used to smooth the spectra is not specified; please provide the filter parameters for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the theoretical prediction is parameter-free and the experimental reduction is an external benchmark.

full rationale

The derivation chain starts from the standard Fabry-Perot transmission formula in Eq. (1), uses the Fresnel reflectance at a silica-air interface in Eq. (4), and expands in powers of cos(delta) in Eqs. (3)-(5). The quarter-wave cancellation is a mathematical consequence of shifting the cavity phase by pi, and the residual error in Eqs. (9)-(11) is computed from tabulated refractive indices and the experimental choice lambda0 = 1.6 um, not from fitted parameters. The measured reduction from 0.34 dB to 0.04 dB is an external experimental result, and no parameter is retrofitted from that data to make the theory match. There are no load-bearing self-citations: all cited works are standard external references on Fabry-Perot interferometry, coupling strategies, and suspended couplers. The manuscript does mention, in Section 5, a control analysis 'performed in which each single measurement was averaged with the subsequent one, mimicking the averaging effect of the CDM,' but it reports no result for that control; this is an evidentiary gap about whether simple averaging alone explains the improvement, not a circular step. Similarly, the unmodeled gap-dependence of the coupling coefficient and the observed Y-axis asymmetry are robustness or correctness concerns, not instances of the derivation reducing to its own inputs. The central claim therefore stands on an independent theoretical and experimental footing.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central derivation uses only standard Fresnel reflectance and the plane-parallel FP transmission formula. No free parameters are fitted to the data; λ0 is a design choice. The main assumptions are the single-cavity model and the constancy of the coupling coefficients over the small gap change.

free parameters (3)

design wavelength λ0 = 1.6 µm
Chosen as the center of the 1505-1675 nm sweep. It defines the quarter-wave shift d' = d + λ0/4 and minimizes the residual error across the band. It is a design choice, not fitted to the reported improvement.
Gaussian filter width = not specified
The spectra were smoothed with a Gaussian filter in Section 5; the filter width is not given, so it is an unspecified analysis parameter that affects the reported σFP values.
outlier threshold and truncation = ±2σ, first 700 points
Applied in Section 5 before computing the histograms in Fig. 4. This post hoc restriction could influence the reported standard deviations, so it is an analysis parameter affecting the validation.

assumptions (3)

domain assumption The fiber-air and oxide-air interfaces form a single parallel-plate Fabry-Perot cavity at normal incidence, with no other reflective interfaces contributing.
Used in Section 3 to write Eq. (1). In practice, the chip has multiple interfaces (waveguide facets, oxide layers) that could add extra reflections, but the model ignores them.
domain assumption The transmission coefficients T1 and T2 are constant over the 400 nm gap change.
Required for Eqs. (7)-(9) to cancel exactly. The paper does not verify this experimentally.
standard math The expansion in Eq. (3) may be truncated after the first-order term for the residual estimate.
The second-order term is about 0.018 dB, which the authors state is negligible. This is a standard approximation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Fabry-Perot-Insensitive Edge Coupling for Robust PIC Characterization." pith.science (2026). https://pith.science/paper/B7XIZBDD

@misc{pith2026250608155,
  author       = {Pith},
  title        = {Pith review of: Fabry-Perot-Insensitive Edge Coupling for Robust PIC Characterization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B7XIZBDD}},
  note         = {Machine review of arXiv:2506.08155}
}

read the original abstract

In this work, we investigate the impact of Fabry-Perot (FP) cavities formed between a fiber array and a silicon chip during edge-coupled testing of photonic integrated circuits (PICs). Our results show that subwavelength variations in the distance between the fiber array and the device under test induce coupling efficiency modulations of several tenths of a decibel - challenging current alignment precision and undermining measurement repeatability in wafer-level probing. To address these FP-induced fluctuations, we introduce a complementary dual-measurement technique that shifts the chip-fiber separation by a quarter wavelength, thereby generating phase-opposite modulation artifacts that cancel upon averaging. This approach allows a higher tolerance in probe-to-PIC positioning by substantially eliminating the FP modulation. We also demonstrate how it can be leveraged to enhance repeatability and accurate positioning during alignment, reducing the insertion loss variability between multiple measurements from 0.34 to 0.04 dB. Such precision is essential for emerging applications like quantum photonics, silicon photonic sensors, and high-bandwidth optical interconnects where measurement accuracy directly impacts device performance characterization. Ultimately, our method offers a robust and accurate solution for high-throughput PIC characterization without requiring index-matching gel, which is impractical for wafer-level testing.

Figures

Figures reproduced from arXiv: 2506.08155 by the authors.

**Figure 2.** Experimental setup for the edge coupling measurements. A sweeping tunable laser is [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Comparison of spectra acquired by a single sweep and CDM. Fig 3a show the transmission [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Histograms showing the distribution of measurement errors for single and paired align [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

7 extracted references · 7 canonical work pages

[1]

Abu-Safia, R

H. Abu-Safia, R. Al-Tahtamouni, I. Abu-Aljarayesh, and N. A. Yusuf. Transmission of a gaus- sian beam through a fabry–perot interferometer. Applied Optics, 33(18):3805–3811, 1994

work page 1994
[2]

A manufacturable platform for photonic quantum computing.Nature, pages 1–3, February 2025

Koen Alexander and et al. A manufacturable platform for photonic quantum computing.Nature, pages 1–3, February 2025. Publisher: Nature Publishing Group

work page 2025
[3]

Efficient suspended coupler with loss less than −1.4 dB between si-photonic waveguide and cleaved single mode fiber

Lianxi Jia, Chao Li, Tsung-Yang Liow, and Guo-Qiang Lo. Efficient suspended coupler with loss less than −1.4 dB between si-photonic waveguide and cleaved single mode fiber. Journal of Lightwave Technology, 36(2):239–244, 2018

work page 2018
[4]

Cou- pling strategies for silicon photonics integrated chips [invited]

Riccardo Marchetti, Cosimo Lacava, Lee Carroll, Kamil Gradkowski, and Paolo Minzioni. Cou- pling strategies for silicon photonics integrated chips [invited]. Photonics Research, 7(2):201, 2019

work page 2019
[5]

M. Vaughan. The Fabry-Perot Interferometer: History, Theory, Practice and Applications. Routledge, 2017

work page 2017
[6]

Diedrik Vermeulen and Christopher V. Poulton. Optical interfaces for silicon photonic circuits. Proceedings of the IEEE, 106(12):2270–2280, 2018

work page 2018
[7]

Ultrabroadband high coupling efficiency fiber-to-waveguide coupler using si3n4 /SiO2 waveguides on silicon

Tiecheng Zhu, Yiwen Hu, Pradip Gatkine, Sylvain Veilleux, Joss Bland-Hawthorn, and Mario Dagenais. Ultrabroadband high coupling efficiency fiber-to-waveguide coupler using si3n4 /SiO2 waveguides on silicon. IEEE Photonics Journal, 8(5):1–12, 2016. 8

work page 2016

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Abu-Safia, R

H. Abu-Safia, R. Al-Tahtamouni, I. Abu-Aljarayesh, and N. A. Yusuf. Transmission of a gaus- sian beam through a fabry–perot interferometer. Applied Optics, 33(18):3805–3811, 1994

work page 1994

[2] [2]

A manufacturable platform for photonic quantum computing.Nature, pages 1–3, February 2025

Koen Alexander and et al. A manufacturable platform for photonic quantum computing.Nature, pages 1–3, February 2025. Publisher: Nature Publishing Group

work page 2025

[3] [3]

Efficient suspended coupler with loss less than −1.4 dB between si-photonic waveguide and cleaved single mode fiber

Lianxi Jia, Chao Li, Tsung-Yang Liow, and Guo-Qiang Lo. Efficient suspended coupler with loss less than −1.4 dB between si-photonic waveguide and cleaved single mode fiber. Journal of Lightwave Technology, 36(2):239–244, 2018

work page 2018

[4] [4]

Cou- pling strategies for silicon photonics integrated chips [invited]

Riccardo Marchetti, Cosimo Lacava, Lee Carroll, Kamil Gradkowski, and Paolo Minzioni. Cou- pling strategies for silicon photonics integrated chips [invited]. Photonics Research, 7(2):201, 2019

work page 2019

[5] [5]

M. Vaughan. The Fabry-Perot Interferometer: History, Theory, Practice and Applications. Routledge, 2017

work page 2017

[6] [6]

Diedrik Vermeulen and Christopher V. Poulton. Optical interfaces for silicon photonic circuits. Proceedings of the IEEE, 106(12):2270–2280, 2018

work page 2018

[7] [7]

Ultrabroadband high coupling efficiency fiber-to-waveguide coupler using si3n4 /SiO2 waveguides on silicon

Tiecheng Zhu, Yiwen Hu, Pradip Gatkine, Sylvain Veilleux, Joss Bland-Hawthorn, and Mario Dagenais. Ultrabroadband high coupling efficiency fiber-to-waveguide coupler using si3n4 /SiO2 waveguides on silicon. IEEE Photonics Journal, 8(5):1–12, 2016. 8

work page 2016