pith. sign in

arxiv: 2511.09440 · v2 · pith:PXTNZY3Xnew · submitted 2025-11-12 · ⚛️ physics.optics

Tutorial: A practical guide to the alignment of defocused spatial light modulators for fast diffractive neural networks

Guillaume Noetinger , Tim Tuuva , Romain Fleury This is my paper

Pith reviewed 2026-05-21 19:26 UTC · model grok-4.3

classification ⚛️ physics.optics
keywords spatial light modulatordiffractive neural networkalignment procedureoptical computingspatial multiplexingwavefront shapingpixel conjugation
0
0 comments X

The pith

A semi-automatic procedure aligns defocused spatial light modulators at pixel level to enable parallel optical diffractive neural networks.

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

The paper presents a practical alignment method for multiple spatial light modulators used in optical diffractive neural networks. Precise conjugation ensures that each spatial channel applies the identical physical transformation to its portion of the input. This setup supports spatial multiplexing so that hundreds of inputs can be processed at once. Training speeds up and noise drops because results can be averaged across channels. The procedure is described as semi-automatic and scalable to any number of modulators.

Core claim

The authors establish that a semi-automatic alignment procedure achieves pixel-level conjugation of defocused spatial light modulators. This conjugation allows construction of an optical diffractive neural network that multiplexes input data across spatial channels, processes hundreds of inputs simultaneously, reduces training times, and lowers experimental noise through spatial averaging.

What carries the argument

Semi-automatic alignment procedure that produces pixel-level conjugation between defocused spatial light modulators.

If this is right

  • The optical setup processes hundreds of inputs simultaneously.
  • Training times decrease because of spatial multiplexing of the data.
  • Experimental noise is reduced by averaging results across channels.
  • The alignment method works for any number of spatial light modulators.
  • The same procedure applies to wavefront shaping experiments that need exact SLM conjugation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This alignment approach could support scaling optical networks to larger input sizes without increasing training duration.
  • Consistent channel behavior might allow direct comparison of optical results with electronic neural network baselines in hybrid systems.
  • The method could be extended to maintain alignment during operation when environmental factors shift the optical path.

Load-bearing premise

Precise alignment is required so that every spatial channel performs exactly the same physical operation and the network can learn consistently.

What would settle it

If identical inputs placed in different spatial channels produce inconsistent network outputs after alignment, the claim of successful pixel-level conjugation would be refuted.

Figures

Figures reproduced from arXiv: 2511.09440 by Guillaume Noetinger, Romain Fleury, Tim Tuuva.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: b. This precision can be attributed to the pre￾cision of the ellipse fitting algorithm. After calibration, the speckle patterns are on average 0.8 pixels apart, com￾pared to 2.5 pixels before calibration. As shown on the histogram 4.c, more than 90% of them are below 1.5 pix￾els, which is below the diffraction limit, resulting in simi￾lar speckle-like outputs. Consequently, after calibration, the sum of ea… view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 6
Figure 6. Figure 6: The reflection from the reflective layer of the [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11 [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12 [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗
read the original abstract

The conjugation of multiple spatial light modulators (SLMs) enables the construction of optical diffractive neural networks (DNNs). To accelerate training, which is limited by the low refresh rate of SLMs, spatial multiplexing of the input data across different spatial channels is possible, maximizing the number of available spatial degrees of freedom (DoFs). Precise alignment is required in order to ensure that the same physical operation is performed across each channel and thus the learning operation of the network. We present a semi-automatic procedure for this experimentally challenging alignment resulting in a pixel-level conjugation. It is scalable to any number of SLMs and may be useful in wavefront shaping setups where precise conjugation of SLMs is required, e.g. for the control of optical waves in phase and amplitude. The resulting setup functions as an optical DNN capable of processing hundreds of inputs simultaneously, thereby reducing training times and experimental noise through spatial averaging. We further present a characterization of the setup and an alignment method.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript presents a semi-automatic experimental procedure for aligning multiple defocused spatial light modulators (SLMs) to achieve pixel-level conjugation. This enables spatial multiplexing in optical diffractive neural networks (DNNs), allowing parallel processing of hundreds of inputs to accelerate training and reduce experimental noise through averaging. The work includes a characterization of the resulting setup and discusses applicability to wavefront shaping.

Significance. If the described alignment achieves consistent pixel-level conjugation across channels, the approach would enable practical, high-throughput optical DNN implementations that leverage spatial degrees of freedom to mitigate the low refresh rates of SLMs. As a tutorial with a scalable procedure, it could provide a valuable methodological resource for experimental groups working on diffractive optics and optical computing, particularly if quantitative validation of alignment precision and network performance is added.

major comments (2)
  1. [Characterization section] Characterization section: the manuscript describes the alignment procedure and setup but provides no quantitative metrics (e.g., measured pixel registration error, conjugation fidelity, or standard deviation across channels) to support the central claim of pixel-level conjugation and its enabling of consistent diffractive operations.
  2. [The section on the resulting optical DNN] The section on the resulting optical DNN: the claim that the setup processes hundreds of inputs simultaneously with reduced training time and noise via spatial averaging is stated without supporting data on throughput, noise reduction factors, or comparison to non-multiplexed operation.
minor comments (2)
  1. Figure captions should more explicitly label the alignment steps and indicate which panels correspond to before/after conjugation to improve clarity for readers following the tutorial.
  2. The abstract and introduction could include a brief forward reference to the specific quantitative characterization results that will be presented later in the manuscript.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify how to better support the central claims of the work. We address each major comment below and indicate the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: [Characterization section] Characterization section: the manuscript describes the alignment procedure and setup but provides no quantitative metrics (e.g., measured pixel registration error, conjugation fidelity, or standard deviation across channels) to support the central claim of pixel-level conjugation and its enabling of consistent diffractive operations.

    Authors: We agree that the characterization section would benefit from explicit quantitative metrics. The manuscript currently focuses on describing the semi-automatic alignment procedure and its scalability, but does not report measured values such as pixel registration error or conjugation fidelity. We will add these metrics in the revised version, for example by quantifying alignment precision through test patterns, interference measurements, and standard deviations across multiple channels. This will directly support the claim of pixel-level conjugation. revision: yes

  2. Referee: [The section on the resulting optical DNN] The section on the resulting optical DNN: the claim that the setup processes hundreds of inputs simultaneously with reduced training time and noise via spatial averaging is stated without supporting data on throughput, noise reduction factors, or comparison to non-multiplexed operation.

    Authors: The manuscript presents the capability for simultaneous processing of hundreds of inputs as a direct consequence of the spatial multiplexing enabled by the alignment procedure. While the principle of noise reduction via averaging is outlined, we acknowledge that specific quantitative data on throughput, noise reduction factors, and direct comparisons are not included. We will add a short subsection with supporting measurements or simulations demonstrating these aspects in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper is explicitly a tutorial on an experimental semi-automatic alignment procedure for defocused SLMs to enable multiplexed optical DNNs. It contains no mathematical derivations, equations, fitted parameters, or predictions that could reduce to inputs by construction. The alignment method and characterization are presented as practical steps with direct experimental support, independent of any self-citation chains or self-definitional loops. The central claim about processing hundreds of inputs in parallel follows from the described scalable procedure rather than any circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work rests on standard assumptions from optics about phase modulation, conjugation, and diffraction; no free parameters, new axioms, or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption SLMs can be used to impose controlled phase shifts on incident light for diffractive computations.
    Invoked implicitly when describing the construction of optical DNNs via conjugation of multiple SLMs.
  • domain assumption Pixel-level conjugation ensures identical physical operations across multiplexed spatial channels.
    Central to the claim that alignment enables consistent learning operations.

pith-pipeline@v0.9.0 · 5704 in / 1344 out tokens · 67229 ms · 2026-05-21T19:26:00.810908+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    To improve detection efficiency, a set of concentric circles with alternating retardance of 0 or 2πis displayed on the SLM instead of a single circle

    for fine-tuning the alignment. To improve detection efficiency, a set of concentric circles with alternating retardance of 0 or 2πis displayed on the SLM instead of a single circle. The image of a dark disk is observed, it is easier to fit than a circle. The corresponding code is available on GitHub [https: //github.com/TTimTT/SLM-alignment] along with a ...

    work page

  2. [2]

    N. e. a. Maalej, Artificial intelligence in- dex report 2024.”, Nature Reviews Physics https://aiindex.stanford.edu/report/. (2024)

    work page 2024

  3. [3]

    Dhar, The carbon impact of artificial intelligence, Na- ture Machine Intelligence2, 423 (2020)

    P. Dhar, The carbon impact of artificial intelligence, Na- ture Machine Intelligence2, 423 (2020)

    work page 2020

  4. [4]

    M. M. Waldrop, The chips are down for moore’s law, Nature530, 144 (2016)

    work page 2016

  5. [5]

    Athale and D

    R. Athale and D. Psaltis, Optical computing: Past and future, Opt. Photon. News27, 32 (2016)

    work page 2016

  6. [6]

    Goodman, Optical computing — an overview, inTop- ical Meeting on Optical Computing(Optica Publishing Group, 1987) p

    J. Goodman, Optical computing — an overview, inTop- ical Meeting on Optical Computing(Optica Publishing Group, 1987) p. MA1

    work page 1987

  7. [7]

    P. L. McMahon, The physics of optical computing, Na- ture Reviews Physics5, 717 (2023)

    work page 2023

  8. [8]

    X. Lin, Y. Rivenson, N. T. Yardimci, M. Veli, Y. Luo, M. Jarrahi, and A. Ozcan, All-optical machine learning using diffractive deep neural networks, Science361, 1004 (2018)

    work page 2018

  9. [9]

    Wetzstein, A

    G. Wetzstein, A. Ozcan, S. Gigan, S. Fan, D. Englund, M. Soljaˇ ci´ c, C. Denz, D. A. B. Miller, and D. Psaltis, Inference in artificial intelligence with deep optics and photonics, Nature (2020)

    work page 2020

  10. [10]

    T. Zhou, X. Lin, J. Wu, Y. Chen, H. Xie, Y. Li, J. Fan, H. Wu, L. Fang, and Q. Dai, Large-scale neuromorphic optoelectronic computing with a reconfigurable diffrac- tive processing unit, Nature Photonics15, 367 (2021)

    work page 2021

  11. [11]

    J. Hu, D. Mengu, D. C. Tzarouchis, B. Edwards, N. En- gheta, and A. Ozcan, Diffractive optical computing in free space, Nature Communications15, 1525 (2024)

    work page 2024

  12. [12]

    Zangeneh-Nejad, D

    F. Zangeneh-Nejad, D. L. Sounas, A. Al` u, and R. Fleury, Analogue computing with metamaterials, Nature Re- views Materials6, 207 (2021)

    work page 2021

  13. [13]

    Wang and R

    H. Wang and R. Piestun, Dynamic 2d implementation of 3d diffractive optics, Optica5, 1220 (2018)

    work page 2018

  14. [14]

    W.Goodman,Introduction to Fourier Optics, 3rd ed

    J. W.Goodman,Introduction to Fourier Optics, 3rd ed. (Roberts & Company, 2005)

    work page 2005

  15. [15]

    S. Chen, Y. Li, Y. Wang, H. Chen, and A. Ozcan, Optical generative models, Nature644, 903 (2025)

    work page 2025

  16. [16]

    R. Chen, Y. Tang, J. Ma, and W. Gao, Scientific comput- ing with diffractive optical neural networks, Advanced In- telligent Systems, Wiley 10.1002/aisy.202300536 (2023)

    work page doi:10.1002/aisy.202300536 2023

  17. [17]

    C. C. Wanjura and F. Marquardt, Fully nonlinear neuro- morphic computing with linear wave scattering, Nature Physics20, 1434 (2024)

    work page 2024

  18. [18]

    Kulce, D

    O. Kulce, D. Mengu, Y. Rivenson, and A. Ozcan, All- optical information-processing capacity of diffractive sur- faces, Light: Science & Applications10, 25 (2021)

    work page 2021

  19. [19]

    Savinson, D

    B. Savinson, D. J. Norris, S. Mishra, and S. Lanthaler, Universality of physical neural networks with multivari- ate nonlinearity, arXiv (2025), 2509.05420

    work page arXiv 2025

  20. [20]

    Zhang and N

    Y. Zhang and N. K. Fontaine, Multi-plane light con- version: A practical tutorial (2023), arXiv:2304.11323 [physics.optics]

    work page arXiv 2023

  21. [21]

    Q. Jia, B. Shi, Y. Zhang, H. Li, X. Li, R. Feng, F. Sun, Y. Cao, J. Wang, C.-W. Qiu, M. Gu, and W. Ding, Par- tially coherent diffractive optical neural network, Optica 11, 1742 (2024)

    work page 2024

  22. [22]

    Yildirim, N

    M. Yildirim, N. U. Dinc, I. Oguz, D. Psaltis, and C. Moser, Nonlinear processing with linear optics, Na- ture Photonics18, 1076 (2024)

    work page 2024

  23. [23]

    J. C. A. Rocha, U. G. B¯ utait˙ e, J. Carpenter, and D. B. Phillips, Self-configuring high-speed multi-plane light conversion (2025), arXiv:2501.14129 [physics.optics]

    work page arXiv 2025

  24. [24]

    Onodera, M

    T. Onodera, M. M. Stein, B. A. Ash, M. M. So- honi, M. Bosch, R. Yanagimoto, M. Jankowski, T. P. McKenna, T. Wang, G. Shvets, M. R. Shcherbakov, L. G. Wright, and P. L. McMahon, Scaling on-chip photonic neural processors using arbitrarily programmable wave propagation, preprint arXiv:2402.17750 (2024)

    work page arXiv 2024

  25. [25]

    O. Lib, R. Shekel, and Y. Bromberg, Building and align- ing a 10-plane light converter, Journal of Physics: Pho- tonics7, 033001 (2025). 8

    work page 2025

  26. [26]

    Gigan, O

    S. Gigan, O. Katz, H. B. de Aguiar, E. R. Andresen, A. Aubry, J. Bertolotti, E. Bossy, D. Bouchet, J. Brake, S. Brasselet, Y. Bromberg, H. Cao, T. Chaigne, Z. Cheng, W. Choi, T. ˇCiˇ zm´ ar, M. Cui, V. R. Curtis, H. De- fienne, M. Hofer, R. Horisaki, R. Horstmeyer, N. Ji, A. K. LaViolette, J. Mertz, C. Moser, A. P. Mosk, N. C. P´ egard, R. Piestun, S. Pop...

    work page 2022

  27. [27]

    K. Y. Bliokh, E. Karimi, M. J. Padgett, M. A. Alonso, M. R. Dennis, A. Dudley, A. Forbes, S. Zahedpour, S. W. Hancock, H. M. Milchberg, S. Rotter, F. Nori, . K. ¨Ozdemir, N. Bender, H. Cao, P. B. Corkum, C. Hern´ andez-Garc´ ıa, H. Ren, Y. Kivshar, M. G. Silveir- inha, N. Engheta, A. Rauschenbeutel, P. Schneeweiss, J. Volz, D. Leykam, D. A. Smirnova, K. R...

    work page 2023

  28. [28]

    Rothe, K

    S. Rothe, K. Wisal, C.-W. Chen, M. Ercan, A. Jesacher, A. D. Stone, and H. Cao, Output beam shaping of a multimode fiber amplifier, arXiv (2024), 2410.23361

    work page arXiv 2024

  29. [29]

    N. U. Dinc, M. Yildirim, I. Oguz, C. Moser, and D. Psaltis, Multicasting optical reconfigurable switch (2024), arXiv:2401.14173 [physics.optics]

    work page arXiv 2024

  30. [30]

    Roadmap on holography, Journal of Optics22, 123002 (2020)

    work page 2020

  31. [31]

    Maurer, A

    C. Maurer, A. Jesacher, S. Bernet, and M. Ritsch-Marte, What spatial light modulators can do for optical mi- croscopy, Laser & Photonics Reviews5, 81 (2011)

    work page 2011

  32. [32]

    Nœtinger, F

    G. Nœtinger, F. Lemoult, and S. M. Popoff, Dynamic structured illumination for confocal microscopy, Opt. Lett.49, 1177 (2024)

    work page 2024

  33. [33]

    Bradski, The OpenCV Library, Dr

    G. Bradski, The OpenCV Library, Dr. Dobb’s Journal of Software Tools (2000])

    work page 2000

  34. [34]

    L. G. Wright, T. Onodera, M. M. Stein, T. Wang, D. T. Schachter, Z. Hu, and P. L. McMahon, Deep physical neural networks trained with backpropagation, Nature 601, 549 (2022)

    work page 2022

  35. [35]

    S. M. Popoff, G. Lerosey, R. Carminati, M. Fink, A. C. Boccara, and S. Gigan, Measuring the transmission ma- trix in optics: An approach to the study and control of light propagation in disordered media, Phys. Rev. Lett. 104, 100601 (2010)

    work page 2010

  36. [36]

    Momeni, B

    A. Momeni, B. Rahmani, B. Scellier, L. G. Wright, P. L. McMahon5, C. C. Wanjura, Y. Li, A. Skalli, N. G. Berloff, T. Onodera, I. Oguz, F. Morichetti, P. del Hougne, M. Le Gallo, A. Sebastian, A. Mirho- seini, C. Zhang, D. Markovic´ c, D. Brunner, C. Moser, S. Gigan, F. Marquardt, A. Ozcan, J. Grollier, A. J. Liu, D. Psaltis, A. Al` u, and R. Fleury, Train...

    work page 2025

  37. [37]

    Momeni, B

    A. Momeni, B. Rahmani, M. Mall´ ejac, P. del Hougne, and R. Fleury, Backpropagation-free training of deep physical neural networks, Science382, 1297 (2023)

    work page 2023

  38. [38]

    F. Xia, K. Kim, Y. Eliezer, L. Shaughnessy, S. Gigan, and H. Cao, Deep learning with passive optical nonlinear mapping, ArXiv (2023)

    work page 2023

  39. [39]

    Jullien, Spatial light modulators, Photoniques (2020)

    A. Jullien, Spatial light modulators, Photoniques (2020)

    work page 2020

  40. [40]

    Marco, A

    D. Marco, A. Vargas, M. del Mar S´ anchez-L´ opez, and I. Moreno, Measuring the spatial deformation of a liquid- crystal on silicon display with a self-interference effect, Opt. Lett.45, 4480 (2020)

    work page 2020

  41. [41]

    G. Bold, T. Barnes, J. Gourlay, R. Sharples, and T. Haskell, Practical issues for the use of liquid crystal spatial light modulators in adaptive optics, Optics Com- munications148, 323 (1998)

    work page 1998

  42. [42]

    Moser, M

    S. Moser, M. Ritsch-Marte, and G. Thalhammer, Model- based compensation of pixel crosstalk in liquid crystal spatial light modulators, Opt. Express27, 25046 (2019)

    work page 2019

  43. [43]

    Mertz,Introduction to Optical Microscopy, 2nd ed

    J. Mertz,Introduction to Optical Microscopy, 2nd ed. (Cambridge University Press, 2019)

    work page 2019

  44. [44]

    Popoff, Setting up a DMD/SLM: Aberration effects, www.wavefrontshaping.net (2016)

    S. Popoff, Setting up a DMD/SLM: Aberration effects, www.wavefrontshaping.net (2016)

    work page 2016

  45. [45]

    Popoff, G

    S. Popoff, G. Lerosey, A. C. Boccara, and S. Gigan, Image transmission through an opaque material, Nature Com- munications1, 81 (2010)

    work page 2010

  46. [46]

    J. S. Heung-Chang Lee, Symba: Symmetric backpropagation-free contrastive learning with forward-forward algorithm for optimizing convergence, arXiv:2303.08418v1 (2023)

    work page arXiv 2023

  47. [47]

    Akiba, S

    T. Akiba, S. Sano, T. Yanase, T. Ohta, and M. Koyama, Optuna: A next-generation hyperparameter optimiza- tion framework, (2019). SUPPLEMENTARY INFORMATIONS A. Axial positioning of the SLM FIG. 6.Optical scheme for the4fsystem conjugating the two modulators.TheµDisplay image fromL 1 −L 2 is indicated with a solid orange line at the left of the optical syst...

    work page 2019