arxiv: 2604.18160 · v1 · submitted 2026-04-20 · ⚛️ physics.optics · cond-mat.dis-nn

Recognition: unknown

Programmable recirculating bricks mesh architecture for photonic neural networks

Jacek Gosciniak

Authors on Pith no claims yet

Pith reviewed 2026-05-10 04:06 UTC · model grok-4.3

classification ⚛️ physics.optics cond-mat.dis-nn

keywords photonic neural networksprogrammable photonic processorswaveguide meshrecirculating brickssingular value decompositionself-calibrationphotonic integrated circuits

0 comments

The pith

A recirculating bricks mesh lets one photonic processor handle crossbars, interference circuits, and SVD after reprogramming.

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

The paper introduces a recirculating bricks mesh architecture for photonic neural networks. This design creates a programmable optical system that can be reconfigured to perform different tasks on the same hardware. It demonstrates that the mesh can switch between a crossbar network and optical interference circuits with variable structures suitable for singular value decomposition. The architecture also incorporates a monitoring system for power tracking at each point, enabling self-calibration and stabilization through feedback loops. This approach aims to provide flexibility in photonic computing, moving beyond fixed application-specific circuits.

Core claim

A single programmable optical system using a recirculating bricks mesh is capable of performing various functions depending on the requirements. The same network, after being reprogrammed, can perform many different functions, ranging from a crossbar network to optical interference circuits with variable structures, which can then be subjected to Singular Value Decomposition. The bricks mesh also provides a foundation for a monitoring system that tracks power in each location of the circuit for self-calibration and stabilization using a feedback loop.

What carries the argument

The recirculating bricks mesh, a reconfigurable waveguide architecture that supports multiple optical functions through reprogramming.

Load-bearing premise

The recirculating bricks mesh can be fabricated and controlled in real photonic integrated circuits with acceptable losses, crosstalk, and reconfiguration speed to enable multi-function reprogramming and self-calibration.

What would settle it

Fabrication and testing of the mesh showing successful reprogramming between a crossbar and an SVD-applicable interference circuit with measured losses and crosstalk below operational thresholds.

read the original abstract

General-purpose programmable photonic processors are considered a crucial technology because they combine the ultra high-speed, massive bandwidth, and energy efficiency of light-based computing with the flexibility of software-defined hardware. Unlike application-specific photonic integrated circuits (ASPIC) designed for one task, these processors use reconfigurable waveguide meshes to implement various functions, such as switching, filtering, or AI computation, on a single chip, allowing for rapid prototyping and versatile, on-demand hardware redefinition. Here we report a recirculating bricks mesh architecture that can be easily implemented in photonic neural networks. It will be shown that a single programmable optical system is capable of performing various functions depending on the requirements. In particular, we will show that the same network, after being reprogrammed, can perform many different functions, ranging from a crossbar network to optical interference circuits with variable structures, which can then be subjected to Singular Value Decomposition. Furthermore, the "bricks" mesh serves as an excellent foundation for implementing a monitoring system capable of monitoring the power in each location of the circuit and, subsequently, sel-fcalibrating and stabilizing the circuit using a feedback loop.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The recirculating bricks mesh is framed as a flexible photonic processor for multiple functions including SVD, but the paper supplies no simulations, equations, or error analysis to support any of the claims.

read the letter

This paper sketches a recirculating bricks mesh architecture for programmable photonic neural networks. The central pitch is that one chip can be reprogrammed on demand to act as a crossbar, as variable-structure interference circuits that then undergo singular value decomposition, and as a self-calibrating system with power monitoring at every point. That is the main takeaway: an architectural variant on existing waveguide meshes that emphasizes recirculation for easier feedback and multi-use flexibility, but presented entirely as future work with no supporting calculations or data.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a recirculating bricks mesh architecture for programmable photonic integrated circuits in neural networks. It claims that a single reconfigurable optical mesh can be reprogrammed on demand to implement multiple distinct functions, including crossbar networks and variable-structure optical interference circuits that can then be used for singular value decomposition (SVD), while also incorporating a power-monitoring system for self-calibration and stabilization via feedback loops.

Significance. If the architecture can be realized with quantified performance bounds, it would offer a versatile hardware substrate for general-purpose photonic processors, reducing reliance on application-specific designs and enabling on-chip prototyping of diverse optical computing tasks such as switching, filtering, and matrix operations central to photonic AI.

major comments (2)

[Abstract] Abstract: The central claim that the same mesh can be reprogrammed from a crossbar topology to variable-structure interference circuits suitable for SVD is stated without any supporting equations, transfer-matrix derivations, simulations, or tolerance analysis. No bound is provided on how per-brick insertion loss, waveguide crosstalk, or phase drift accumulate under reconfiguration, even though SVD accuracy is known to degrade rapidly under small matrix perturbations.
[Abstract] Abstract and architecture description: The self-calibration claim via integrated power monitoring and feedback is presented prospectively but without a concrete model of the monitoring network, the feedback-loop dynamics, or how calibration corrects for the very losses and crosstalk that would affect SVD fidelity in the reconfigured states.

minor comments (1)

[Abstract] Abstract contains a typographical error: 'sel-fcalibrating' should read 'self-calibrating'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and the opportunity to clarify and strengthen our manuscript on the recirculating bricks mesh architecture. We address the major comments point by point below.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim that the same mesh can be reprogrammed from a crossbar topology to variable-structure interference circuits suitable for SVD is stated without any supporting equations, transfer-matrix derivations, simulations, or tolerance analysis. No bound is provided on how per-brick insertion loss, waveguide crosstalk, or phase drift accumulate under reconfiguration, even though SVD accuracy is known to degrade rapidly under small matrix perturbations.

Authors: We agree that the abstract presents the claim at a high level without derivations or analysis. The body of the manuscript provides a qualitative description of the architecture and its reconfiguration capability but does not include explicit transfer-matrix derivations, numerical simulations of the crossbar-to-interference-circuit mapping, or quantitative tolerance bounds on loss, crosstalk, and phase drift for SVD fidelity. In the revised manuscript we have added these elements: transfer-matrix equations for brick-level reconfiguration, basic numerical examples of SVD on the reconfigured mesh, and first-order bounds on error accumulation (e.g., relating per-brick loss to singular-value perturbation). A full statistical tolerance study remains outside the present scope and is noted as future work. revision: yes
Referee: [Abstract] Abstract and architecture description: The self-calibration claim via integrated power monitoring and feedback is presented prospectively but without a concrete model of the monitoring network, the feedback-loop dynamics, or how calibration corrects for the very losses and crosstalk that would affect SVD fidelity in the reconfigured states.

Authors: We concur that the self-calibration discussion is prospective and lacks a concrete model. The original manuscript mentions power monitoring at brick locations and feedback stabilization but supplies neither the monitoring network topology, loop dynamics, nor quantitative demonstration that calibration mitigates the same impairments that degrade SVD. The revised version now includes a schematic of the integrated photodetector network, a simple proportional-integral feedback model, and a brief analysis showing how power-based correction reduces effective crosstalk in reconfigured states. Detailed closed-loop simulations of SVD accuracy post-calibration are noted as beyond the current scope. revision: yes

Circularity Check

0 steps flagged

No circularity: architecture claims are descriptive without self-referential derivations

full rationale

The manuscript presents a conceptual recirculating bricks mesh for programmable photonic neural networks, emphasizing reconfigurability across functions such as crossbar networks and variable interference circuits for SVD. No equations, fitted parameters, or derivation steps appear in the abstract or described content that reduce by construction to inputs. Claims rely on physical feasibility of fabrication and control rather than tautological definitions, self-citation load-bearing premises, or renamed empirical patterns. The derivation chain is self-contained as forward-looking engineering description without the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the architecture is described at conceptual level without quantitative details.

pith-pipeline@v0.9.0 · 5491 in / 954 out tokens · 30199 ms · 2026-05-10T04:06:50.897770+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

31 extracted references · 4 canonical work pages

[1]

Brain-inspired computing needs a master plan ,

A. Mehonic, A. J. Kenyon, “Brain-inspired computing needs a master plan ,” Nature 604, 255–260 (2022)

2022
[2]

Neuromorphic computing at scale,

D. Kudithipudi, et al., “Neuromorphic computing at scale,” Nature 637, 801–812 (2025)

2025
[3]

Compute Trends Across Three Eras of Machine Learning

J. Sevilla, L. Heim, A. Ho, T. Besiroglu, M. Hobbhahn, P . Villalobos, “Compute Trends Across Three Eras of Machine Learning” in 2022 International Joint Conference on Neural Networks (IJCNN), Padua, Italy, pp. 1–8 (2022)

2022
[4]

Large-scale optical neural networks based on photoelectric multiplication,

R. Hamerly, L. Bernstei, A. Sludds, M. Soljačić, D. Englund, “Large-scale optical neural networks based on photoelectric multiplication,” Phys. Rev. X 9, 021032 (2019)

2019
[5]

Single-chip photonic deep neural network with forward-only training,

S. Bandyopadhyay, R. Hamerly, A. Sludds, N. Harris, S. Krastanov, D. Bunandar, M. Streshinsky, M. Hochberg, D. Englund, “Single-chip photonic deep neural network with forward-only training,” Nat. Photonics 18, 1335-1343 (2024)

2024
[6]

Integrated photonic neural network with on-chip backpropagation training,

F. Ashtiani, M. H. Idjadi, K. Kim, “Integrated photonic neural network with on-chip backpropagation training,” Nature 651, 927–932 (2026)

2026
[7]

Programmable photonic circuits,

W. Bogaerts, D. Pérez, J. Capmany, D. A. B. Miller, J. Poon, D. Englund, F. Morichetti, A. Melloni, “Programmable photonic circuits,” Nature 586 (7828), 207-216 (2020)

2020
[8]

Field -programmable photonic arrays,

D. Perez, I. Gasulla, J. Capmany, “Field -programmable photonic arrays,” Opt. Express 26 (21), 27265–27278 (2018)

2018
[9]

A programmable topological photonic chip,

T. Dai, A. Ma, J. Mao, Y . Ao, X. Jia, Y . Zheng, CH. Zhai, Y . Yang, Z. Li, B. Tang, J. Luo, B. Zhang, X. Hu, A. Gong, I. Wang, “A programmable topological photonic chip,” Nat. Materials 23, 928–936 (2024)

2024
[10]

Toward the information-theoretic limit of programmable photonics,

R. Hamerly, J. R. Basani, A. Sludds, S. K. Vadlamani, D. Englund, “Toward the information-theoretic limit of programmable photonics,” APL Photon. 10, 110803 (2025)

2025
[11]

Partial coherence enhances parallelized photonic computing,

B. Dong et al., “Partial coherence enhances parallelized photonic computing,” Nature 632, 55–62 (2024)

2024
[12]

Parallel convolutional processing using an integrated photonic tensor core ,

J. Feldmann, N. Youngblood , M. Karpov, H. Gehring, X. Li, M. Stappers, M. Le Gallo, X. Fu, A. Lukashchuk, A. S. Raja, J. Liu, C. D. Wright, A. Sebastian, T. J. Kippenberg, W. H. P . Pernice , H. Bhaskaran, “Parallel convolutional processing using an integrated photonic tensor core ,” Nature 589, 52–58 (2021)

2021
[13]

A coherent photonic crossbar for scalable universal linear optics,

G. Giamougiannis, A. Tsakyridis, Y. Ma, A . Totović, M. Moralis-Pegios, D. Lazovsky, N. Pleros, “A coherent photonic crossbar for scalable universal linear optics,” J. Lightwave Technol. 41 (8), 2425- 2442 (2023)

2023
[14]

On- chip 1 TOPS Hyperdimensional Photonic Tensor Core using a WDM Silicon Photonic Coherent Crossbar,

S. Kovaios, I. Roumpos, A. Tsakyridis, M. Moralis-Pegios, D. Lazovsky, K. Vyrsokinos, N. Pleros, “On- chip 1 TOPS Hyperdimensional Photonic Tensor Core using a WDM Silicon Photonic Coherent Crossbar,” J. Lightwave Technol. 43 (18), 8799 - 8805 (2025)

2025
[15]

Shifted rectangular mesh architecture for programmable photonics,

J. Gosciniak, “Shifted rectangular mesh architecture for programmable photonics,” arXiv, arXiv:2510.17307 (2025)

work page arXiv 2025
[16]

Recirculating Quantum Photonic Networks for Fast Deterministic Quantum Information Processing,

E. Grovn, M. Bundgaard-Nielsen, J. Mork, D. Englund, M. Heuck, “Recirculating Quantum Photonic Networks for Fast Deterministic Quantum Information Processing,” arXiv, arXiv:2602.11033 (2026)

work page arXiv 2026
[17]

Programmable recirculating bricks mesh architecture for quantum photonics,

J. Gosciniak, “Programmable recirculating bricks mesh architecture for quantum photonics,” arXiv, arXiv:2604.01369 (2026)

work page arXiv 2026
[18]

Gosciniak, “Automated control strategy for setting and stabilization of photonic circuits,“ arXiv, arXiv:2510.22737 (2025)

J. Gosciniak, “Automated control strategy for setting and stabilization of photonic circuits,“ arXiv, arXiv:2510.22737 (2025)

work page arXiv 2025
[19]

Optimal design for universal multiport interferometers,

W. R. Clements, P . C. Humphreys, B. J. Metcalf, W. S. Kolthammer, I. A. Walmsley, “Optimal design for universal multiport interferometers,” Optica 3 (12), 1460-1465 (2016)

2016
[20]

Further compactifying linear optical unitaries,

B. A. Bell, I. A. Walmsley, “Further compactifying linear optical unitaries,” APL Photon. 6, 070804 (2021). 10

2021
[21]

A logical calculus of the ideas immanent in nervous activity,

W. S. McCulloch, W . Pitts, “A logical calculus of the ideas immanent in nervous activity,” The bulletin of mathematical biophysics, 5 (4), 115-133 (1943)

1943
[22]

Deep learning with coherent nanophotonic circuits ,

Y. Shen, N . C. Harris, S . Skirlo, M . Prabhu, T. Baehr-Jones, M . Hochberg, X . Sun, S . Zhao, H . Larochelle, D . Englund, M. Soljacic, “Deep learning with coherent nanophotonic circuits ,” Nat. Photonics 11, 441–446 (2017)

2017
[23]

Measuring, processing, and generating partially coherent light with self-configuring optics,

Ch. Roques-Carmes, S. Fan, D. A. B. Miller , “Measuring, processing, and generating partially coherent light with self-configuring optics,” Light Sci. Appl. 13, 260 (2024)

2024
[24]

Large-scale photonic chiplet Taichi empowers 160-TOPS/W artificial general intelligence,

Z. Xu, T. Zhou, M. Ma, Ch. Deng, Q. Dai2, L. Fang, “Large-scale photonic chiplet Taichi empowers 160-TOPS/W artificial general intelligence,” Science 384 (6692), 202-209 (2024)

2024
[25]

Bistable all-optical devices based on nonlinear epsilon-near-zero (ENZ) materials,

J. Gosciniak, Z. Hu, M. Thomaschewski, V. J. Sorger, J. B. Khurgin, “Bistable all-optical devices based on nonlinear epsilon-near-zero (ENZ) materials,” Laser Photonics Rev. 17 (4), 2200723 (2023)

2023
[26]

Experimentally realized in situ backpropagation for deep learning in photonic neural networks,

S. Pai, Z. Sun, T. W. Hughes2, T. Park, B. Bartlett2, I. A. D. Williamson, M. Minkov, M. Milanizadeh, N. Abebe, F. Morichetti, A. Melloni, S. Fan, O. Solgaard, D. A. B. Miller, “Experimentally realized in situ backpropagation for deep learning in photonic neural networks,” Science 380 (6643), 398-404 (2023)

2023
[27]

Universal photonic artificial intelligence acceleration ,

S. R. Ahmed et al., “Universal photonic artificial intelligence acceleration ,” Nature 640, 368–374 (2025)

2025
[28]

Training of photonic neural networks through in situ backpropagation and gradient measurement,

T. W. Hughes, M . Minkov, Y. Shi, S . Fan, “ Training of photonic neural networks through in situ backpropagation and gradient measurement,” Optica 5 (7), 864-871 (2018)

2018
[29]

Integrated Bolometric Photodetectors Based on Transparent Conductive Oxides from Near-to Mid-Infrared Wavelengths,

J. Gosciniak, “Integrated Bolometric Photodetectors Based on Transparent Conductive Oxides from Near-to Mid-Infrared Wavelengths,” Adv. Photonics Res. 5 (2), 2300261 (2024)

2024
[30]

Transparent conductive oxides as a material platform for a realization of all-optical photonic neural networks,

J. Gosciniak, J. B. Khurgin, “Transparent conductive oxides as a material platform for a realization of all-optical photonic neural networks,” Sci. Rep. 15, 24977 (2025)

2025
[31]

Gosciniak, J

J. Gosciniak, J. B. Khurgin , ”Schottky photodetectors with transparent conductive oxides for photonic integrated circuits,” ACS Photonics 11 (3), 1137-1146 (2024)

2024