REVIEW 3 major objections 6 minor 1 cited by
A low-loss, 24-mode laser-written universal photonic processor in a glass-based platform
T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash
Pith's one-line read A 24-mode universal photonic processor written into glass by femtosecond lasers reproduces 2000 Haar-random unitaries with 99.7% average amplitude fidelity, at under 10 W total electrical power.
desk verdict A genuine fabrication step — first 24-mode FLW universal processor — but the 99.7% amplitude fidelity is undefined and may compare only moduli, so the central unitary-implementation claim is not yet verifiable. 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 object is the 24-mode mesh of Mach–Zehnder interferometers written into glass, each interferometer controlled by two thermal phase shifters. Around the waveguides, 60-µm-deep and 1-mm-long isolation trenches create suspended bridge structures that suppress thermal crosstalk, while a two-metal layer of chromium heaters and copper interconnections delivers phase shifts efficiently. The performance claim rests on the calibration model: 14,952 fitted parameters (576 static phases, 552 coupler splitting ratios, and 13,824 thermal-crosstalk coefficients) trained on 30,000 uniformly random phase configurations. This model converts desired unitaries into heater powers and is what makes the 99.7% average amplitude fidelity on held-out Haar-random unitaries possible.
What would settle it
Take the same chip and measure amplitude fidelity on target unitaries deliberately chosen from a different distribution than the training data, for instance phase-shifter configurations with all heaters near their maximum 45 mW power, or spatially correlated settings, and check whether the fidelity stays near 99.7%. A significant drop would show the calibration model is fitted to the training distribution rather than a true model of the device.
Extended reading notes
Core claim
The central discovery is a 24-mode universal photonic processor realized in glass by femtosecond laser writing, which the authors state is the most complex processor of its kind reported to date. The device is a mesh of Mach–Zehnder interferometers with 552 directional couplers and 576 thermal phase shifters, optimized for 925 nm light. Thanks to suspended-bridge isolation trenches and a two-metal heater process, the average fiber-to-fiber insertion loss is 4.35 dB and the power consumption stays below 10 W for all tested transformations, including Haar-random unitaries, switching operations, and random phase settings. After a machine-learning calibration against 30,000 random phase configurations, the chip implements 2000 Haar-random unitaries with an average amplitude fidelity of 99.7%. The paper concludes that these results establish femtosecond laser writing as a scalable platform for integrated quantum photonic circuits.
Load-bearing premise
The calibration model trained on 30,000 random phase settings must predict the device's behavior for arbitrary settings outside that sample, including settings unlike anything it saw during training.
Editorial extensions
If this is right
- At 24 modes, this is the largest universal photonic processor demonstrated in the femtosecond-laser-writing platform, enabling circuits beyond the 20-mode devices previously reported.
- Average fiber-to-fiber insertion loss of 4.35 dB at 925 nm means the device can be driven directly by InGaAs quantum-dot single-photon sources without loss-compensating amplification.
- Total electrical power below 10 W for every tested transformation makes the whole processor manageable with a simple thermoelectric cooler, simplifying packaging and operation.
- The calibration approach, a machine-learning model trained on 30,000 random phase configurations, yields 99.7% average amplitude fidelity on 2000 held-out Haar-random unitaries, showing the processor can be programmed for essentially arbitrary linear-optical transformations.
- Curved isolation trenches and the two-metal heater process reduce both crosstalk and thermal power, pointing a path toward denser and larger meshes in glass.
Reading between the lines
- This suggests the same calibration approach could scale to larger meshes, since the parameter count grows quadratically with mode number; testing it on a 30- or 40-mode chip would be the natural check.
- The low loss and low power budgets are exactly what a multi-photon experiment needs; although the paper reports only classical-light characterization, a boson-sampling or multi-photon interference run would directly exercise the claimed capability.
- Amplitude fidelity alone does not fully characterize a quantum processor; measuring phase fidelity via two-photon interference fringes or process tomography would be a stronger test of the calibration model.
- The explicit thermal-crosstalk matrix could be used as a design tool for future chips, for example to choose heater spacing and trench geometry that minimize the cross-talk coefficients before fabrication.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This manuscript reports the fabrication and characterization of a 24-mode universal photonic processor (UPP) realized by femtosecond laser writing in glass, optimized for 925 nm operation. The authors describe a two-metal microheater integration process, suspended waveguides with isolation trenches, and a calibration procedure based on 30,000 measured random phase configurations. They report average fiber-to-fiber insertion losses of 4.35 dB, total electrical power below 10 W for all tested transformations, and an average amplitude fidelity of 99.7% over 2000 held-out Haar-random unitaries. The paper concludes that the device represents the most complex UPP demonstrated to date in this platform.
Significance. If the reported results are confirmed, this work would be a significant advance for femtosecond-laser-written programmable photonic circuits, demonstrating a 24-mode reconfigurable processor with low insertion loss and low electrical power, together with a calibration procedure that generalizes to randomly chosen target unitaries. The held-out test on 2000 Haar-random unitaries is a commendable generalization check and weakens circularity concerns. The hardware innovations, including suspended waveguides and a two-metal lithography process, are clearly described and credible. The main weakness is the undefined fidelity metric, which is load-bearing for the central quantitative claim and must be clarified before the results can be fully assessed.
major comments (3)
- [Section 3.2, Fig. 4] The central quantitative claim—an average amplitude fidelity of 99.7% on 2000 Haar-random unitaries—rests on a metric that is never defined in the manuscript. No formula is given, and the text does not describe how the complex matrix elements of the implemented unitary are measured. The caption of Fig. 4b, 'Amplitudes of a target and measured Haar-random unitary transformation,' suggests that only the moduli |U_ij| are compared. If that is the case, the metric is insensitive to relative phase errors, which are precisely what determines multi-photon interference and the actual content of implementing a unitary transformation. The authors must provide the exact fidelity definition, state whether it incorporates phase information, and specify the measurement protocol, including any interferometric phase reconstruction, that supports the claim.
- [Section 3.2] The paper reports no uncertainty or distribution for the fidelity. The scatter plot in Fig. 4a shows variation across the 2000 tested unitaries, but no standard deviation, minimum, maximum, or experimental error bars are provided. Without these, the reader cannot judge whether the 99.7% average is a robust summary or is dominated by a subset of high-fidelity points. Please report the full distribution and the estimated measurement uncertainty.
- [Section 3.2] The calibration description omits the inverse-model step: how a target unitary matrix is mapped to the 576 phase-shifter control parameters. It is therefore unclear whether the reported fidelity includes errors from this inversion or only from the forward model's predictive accuracy for random phase settings. Since programming the processor requires this mapping, the inverse step is part of the device operation; please specify it and clarify its role in the fidelity evaluation.
minor comments (6)
- [Section 2.2] The manuscript states that the device has 576 microheaters, but a 24-mode triangular mesh would normally contain 552 MZI phase shifters (276 MZIs times 2). The function of the extra 24 heaters is not explained; please clarify the mesh geometry and the role of each heater.
- [Figure 2c] The y-axis of the electrical stability plot is not labeled in the available text; please ensure all axes have readable labels and units.
- [Section 2.1] The mesh architecture (e.g., Reck triangular, Clements rectangular) is not specified. This information is needed to understand the circuit layout, the number of couplers, and the calibration model; please state it.
- [Section 3.2] The training procedure is described only as a 'machine learning model'; please specify the model class (e.g., a physics-based forward model with learned parameters), the loss function used, and how the 13,824 thermal cross-talk coefficients are regularized.
- [Introduction] The claim that this is the 'most complex UPP demonstrated to date' would be more convincing with a quantitative comparison to the 20-mode processor of Ref. [4] in terms of mode count, loss, fidelity, and power.
- [Section 3.1] The text begins 'This sections presents' and should read 'This section presents.'
Circularity Check
No significant circularity: the 99.7% amplitude fidelity is a held-out test of a calibration model, and the cited prior work is methodological rather than load-bearing.
full rationale
The paper's central quantitative claim is the 99.7% amplitude fidelity over 2000 Haar-random unitary matrices after a calibration procedure. The calibration fits a machine-learning model of the device (576 static phases, 552 coupler splitting ratios, and 13,824 thermal crosstalk coefficients) using 30,000 measured unitaries with random phase settings; the reported fidelity is then evaluated on a separate set of 2000 Haar-random target unitaries that were not used in the fit. This is a standard held-out experimental validation, not a prediction forced by construction. The only nearby concern is that the 'amplitude fidelity' metric is never formally defined and appears to compare matrix-element magnitudes rather than full complex unitaries; that is a reporting and verification-quality issue, not circularity. The self-citations in the paper (references [10], [15], [16]) concern prior fabrication techniques for laser-written circuits and microheaters; they are used as methodological background, not as the logical premise for the device's claimed performance. No uniqueness theorem, ansatz, or fitted datum is renamed as a prediction, and no derivation reduces to its own inputs. The observed performance is therefore an independent experimental result within the paper's stated calibration-and-test protocol.
Assumptions & free parameters
free parameters (3)
- thermal crosstalk coefficients =
13,824 (values not listed)
- directional coupler splitting ratios =
552 (values not listed)
- static phase contributions =
576 (values not listed)
assumptions (3)
- standard math A rectangular mesh of Mach-Zehnder interferometers can implement any NxN unitary transformation
- domain assumption The device response is stable over the calibration and test period, so fitted parameters remain valid
- domain assumption Thermal phase shifters provide a monotonic and repeatable phase change versus dissipated power
Cite this review
Pith. "Pith review of A low-loss, 24-mode laser-written universal photonic processor in a glass-based platform." pith.science (2026). https://pith.science/paper/OJKBF3GZ
@misc{pith2026250501609,
author = {Pith},
title = {Pith review of: A low-loss, 24-mode laser-written universal photonic processor in a glass-based platform},
year = {2026},
howpublished = {\url{https://pith.science/paper/OJKBF3GZ}},
note = {Machine review of arXiv:2505.01609}
}
read the original abstract
We report the fabrication of the first 24-mode universal photonic processor (UPP) realized through femtosecond laser writing (FLW), marking the most complex UPP demonstrated to date. Optimized for quantum dot emission at 925 nm, the device exhibits total insertion losses averaging only 4.35 dB, enabling its direct application in advanced multi-photon quantum experiments. Leveraging the versatility of FLW, we introduce suspended waveguides and precisely engineered 2D and 3D microstructures, significantly enhancing thermal isolation and minimizing power dissipation. As a result, our processor operates efficiently at less than 10 W, requiring only a simple thermo-electric cooler for stable thermal management. The device exhibits exceptional performance after calibration, implementing Haar-random unitary transformations with an amplitude fidelity of 99.7 %. This work establishes FLW-based integrated photonics as a scalable and robust platform for advancing quantum computing, communication, and sensing technologies.
Figures
Figures from the paper (1 more)
Forward citations
Cited by 1 Pith paper
-
Detecting high-dimensional entanglement with simple measurements
A witness method detects high-dimensional Schmidt numbers using only strings of single-qubit Pauli measurements, demonstrated up to 16-dimensional photonic entanglement.
Reference graph
Works this paper leans on
-
[1]
The potential and global outlook of integrated photonics for quantum technologies,
Pelucchi, E., Fagas, G., Aharonovich, I., Englund, D., Figueroa, E., Gong, Q., Hannes, H., Liu, J., Lu, C.-Y., Matsuda, N., et al., “The potential and global outlook of integrated photonics for quantum technologies,” Nature Reviews Physics4(3), 194–208 (2022)
work page 2022
-
[2]
Programmable photonic circuits,
Bogaerts, W., P´ erez, D., Capmany, J., Miller, D. A., Poon, J., Englund, D., Morichetti, F., and Melloni, A., “Programmable photonic circuits,”Nature586(7828), 207–216 (2020)
work page 2020
-
[3]
Linear programmable nanophotonic processors,
Harris, N. C., Carolan, J., Bunandar, D., Prabhu, M., Hochberg, M., Baehr-Jones, T., Fanto, M. L., Smith, A. M., Tison, C. C., Alsing, P. M., et al., “Linear programmable nanophotonic processors,”Optica5(12), 1623–1631 (2018)
work page 2018
-
[4]
20-mode universal quantum photonic processor,
Taballione, C., Anguita, M. C., de Goede, M., Venderbosch, P., Kassenberg, B., Snijders, H., Kannan, N., Vleeshouwers, W. L., Smith, D., Epping, J. P., et al., “20-mode universal quantum photonic processor,” Quantum7, 1071 (2023)
work page 2023
-
[5]
Experimental realization of any discrete unitary operator,
Reck, M., Zeilinger, A., Bernstein, H. J., and Bertani, P., “Experimental realization of any discrete unitary operator,”Physical Review Letters73(1), 58 (1994)
work page 1994
-
[6]
Optimal design for universal multiport interferometers,
Clements, W. R., Humphreys, P. C., Metcalf, B. J., Kolthammer, W. S., and Walmsley, I. A., “Optimal design for universal multiport interferometers,”Optica3(12), 1460–1465 (2016)
work page 2016
-
[7]
High fidelity 12-mode quantum photonic processor oper- ating at InGaAs quantum dot wavelength,
de Goede, M., Snijders, H., Venderbosch, P., Kassenberg, B., Kannan, N., Smith, D., Taballione, C., Epping, J. P., van den Vlekkert, H., and Renema, J. J., “High fidelity 12-mode quantum photonic processor oper- ating at InGaAs quantum dot wavelength,” in [Integrated Photonics Research, Silicon and Nanophotonics], ITu4B–3, Optica Publishing Group (2022)
work page 2022
-
[8]
Carolan, J., Harrold, C., Sparrow, C., Mart´ ın-L´ opez, E., Russell, N. J., Silverstone, J. W., Shadbolt, P. J., Matsuda, N., Oguma, M., Itoh, M., et al., “Universal linear optics,”Science349(6249), 711–716 (2015)
work page 2015
Show all 17 references
-
[9]
Reconfigurable photonics on a glass chip,
Dyakonov, I., Pogorelov, I., Bobrov, I., Kalinkin, A., Straupe, S., Kulik, S., Dyakonov, P., and Evlashin, S., “Reconfigurable photonics on a glass chip,”Physical Review Applied10(4), 044048 (2018)
2018
-
[10]
High- fidelity and polarization-insensitive universal photonic processors fabricated by femtosecond laser writing,
Pentangelo, C., Di Giano, N., Piacentini, S., Arpe, R., Ceccarelli, F., Crespi, A., and Osellame, R., “High- fidelity and polarization-insensitive universal photonic processors fabricated by femtosecond laser writing,” Nanophotonics13(12), 2259–2270 (2024)
2024
-
[11]
Femtosecond laser micromachining for integrated quantum photonics,
Corrielli, G., Crespi, A., and Osellame, R., “Femtosecond laser micromachining for integrated quantum photonics,”Nanophotonics10(15), 3789–3812 (2021)
2021
-
[12]
Symmetric polarization- insensitive directional couplers fabricated by femtosecond laser writing,
Corrielli, G., Atzeni, S., Piacentini, S., Pitsios, I., Crespi, A., and Osellame, R., “Symmetric polarization- insensitive directional couplers fabricated by femtosecond laser writing,”Optics Express26(12), 15101–15109 (2018). 5
2018
-
[13]
Suppression law of quantum states in a 3D photonic fast fourier transform chip,
Crespi, A., Osellame, R., Ramponi, R., Bentivegna, M., Flamini, F., Spagnolo, N., Viggianiello, N., Inno- centi, L., Mataloni, P., and Sciarrino, F., “Suppression law of quantum states in a 3D photonic fast fourier transform chip,”Nature Communications7(1), 10469 (2016)
2016
-
[14]
Reconfigurable continuously-coupled 3D photonic circuit for boson sampling experiments,
Hoch, F., Piacentini, S., Giordani, T., Tian, Z.-N., Iuliano, M., Esposito, C., Camillini, A., Carvacho, G., Ceccarelli, F., Spagnolo, N., et al., “Reconfigurable continuously-coupled 3D photonic circuit for boson sampling experiments,”npj Quantum Information8(1), 55 (2022)
2022
-
[15]
Low power recon- figurability and reduced crosstalk in integrated photonic circuits fabricated by femtosecond laser microma- chining,
Ceccarelli, F., Atzeni, S., Pentangelo, C., Pellegatta, F., Crespi, A., and Osellame, R., “Low power recon- figurability and reduced crosstalk in integrated photonic circuits fabricated by femtosecond laser microma- chining,”Laser & Photonics Reviews14(10), 2000024 (2020)
2020
-
[16]
Toward higher integration density in femtosecond-laser-written programmable photonic circuits,
Albiero, R., Pentangelo, C., Gardina, M., Atzeni, S., Ceccarelli, F., and Osellame, R., “Toward higher integration density in femtosecond-laser-written programmable photonic circuits,”Micromachines13(7), 1145 (2022)
2022
-
[17]
(Visited on 07/02/2025).https://www.polifab.polimi.it/. 6
2025
Reviewed August 16, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.