CoChat
Sign Up
Cited by 255Neural Networks and Reservoir ComputingRandom lasers and scattering mediaAdvanced Memory and Neural Computing

Analysis of Diffractive Optical Neural Networks and Their Integration With Electronic Neural Networks

Open access
Mengu, Deniz, Luo, Yi, Rivenson, Yair, Ozcan, Aydogan·2020·IEEE Journal of Selected Topics in Quantum Electronics
DOI 10.1109/jstqe.2019.2921376PMC PMC7677864
Open this paper in CoChat — free accountOpen PDF

Abstract

Optical machine learning offers advantages in terms of power efficiency, scalability, and computation speed. Recently, an optical machine learning method based on diffractive deep neural networks (D2NNs) has been introduced to execute a function as the input light diffracts through passive layers, designed by deep learning using a computer. Here, we introduce improvements to D2NNs by changing the training loss function and reducing the impact of vanishing gradients in the error back-propagation step. Using five phase-only diffractive layers, we numerically achieved a classification accuracy of 97.18% and 89.13% for optical recognition of handwritten digits and fashion products, respectively; using both phase and amplitude modulation (complex-valued) at each layer, our inference performance improved to 97.81% and 89.32%, respectively. Furthermore, we report the integration of D2NNs with electronic neural networks to create hybrid classifiers that significantly reduce the number of input pixels into an electronic network using an ultra-compact front-end D2NN with a layer-to-layer distance of a few wavelengths, also reducing the complexity of the successive electronic network. Using a five-layer phase-only D2NN jointly optimized with a single fully connected electronic layer, we achieved a classification accuracy of 98.71% and 90.04% for the recognition of handwritten digits and fashion products, respectively. Moreover, the input to the electronic network was compressed by >7.8 times down to 10 × 10 pixels. Beyond creating low-power and high-frame rate machine learning platforms, D2NN-based hybrid neural networks will find applications in smart optical imager and sensor design.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1109/jstqe.2019.2921376

Cites

4 of 4 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Cascading a SIM with an electronic neural network to create a HOENN can overcome the linear transfer function limitation.

    “To overcome this limitation, an effective solution is to cascade a SIM with an electronic neural network for creating a HOENN [116].”

    From Emerging Technologies in Intelligent Metasurfaces: Shaping the Future of Wireless Communications (2026) · cites this paper as [116]

    The Methods section describes cascading a D2NN optical front-end with a digital neural network, and the Results and Discussion section reports that an added electronic layer can compensate for some optical-front-end limitations. The Introduction says the optical network uses linear materials, but the paper does not establish that this hybrid overcomes a linear transfer-function limitation specifically, or evaluate a SIM/HOENN by those names. Page numbers are unavailable in our copy. (LLM, conf=0.82)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 32

  1. [1]Caulfield HJ, Kinser J, and Rogers SK, “ Optical neural networks,” Proc. IEEE, vol. 77 no. 10 pp. 1573– 1583. 10 1989.
  2. [2]Yu FTS and Jutamulia S, Optical Signal Processing, Computing, and Neural Networks, 1st ed. New York, NY, USA: John Wiley & Sons, inc, 1992.
  3. [3]Yu FTS, “ II Optical Neural Networks: Architecture, Design and Models,” in Progress in Optics, vol. 32, Wolf E, Ed. Elsevier, 1993, pp. 61– 144.
  4. [4]Psaltis D and Farhat N, “ Optical information processing based on an associative-memory model of neural nets with thresholding and feedback,” Opt. Lett, vol. 10, no. 2, pp. 98– 100, 2 1985. 19724358
  5. [5]Farhat NH, Psaltis D, Prata A, and Paek E, “ Optical implementation of the Hopfield model,” Appl. Opt, vol. 24, no. 10, pp. 1469– 1475. 5 1985. 18223740
  6. [6]Wagner K and Psaltis D, “ Multilayer optical learning networks,” Appl. Opt, vol. 26, no. 23, pp. 5061– 5076, 12 1987. 20523485
  7. [7]Psaltis D, Brady D, and Wagner K, “ Adaptive optical networks using photorefractive crystals,” Appl. Opt, vol. 27, no. 9, pp. 1752– 1759. 5 1988. 20531647
  8. [8]Psaltis D, Brady D, Gu X-G, and Lin S, “ Holography in artificial neural networks,” Nature, vol. 343, no. 6256, pp. 325– 330, 1 1990. 2300184
  9. [9]Weverka RT, Wagner K, and Saffman M, “ Fully interconnected, two-dimensional neural arrays using wavelength-multiplexed volume holograms,” Opt. Lett, vol. 16, no. 11, pp. 826– 828, 6 1991. 19776798
  10. [10]Javidi B, Li J, and Tang Q, “ Optical implementation of neural networks for face recognition by the use of nonlinear joint transform correlators,” Appl. Opt, vol. 34, no. 20, pp. 3950– 3962, 7 1995. 21052218
  11. [11]Shen Y, “ Deep learning with coherent nanophotonic circuits,” Nat. Photonics, vol. 11, no. 7, pp. 441– 446, 7 2017.
  12. [12]Bueno J, “ Reinforcement learning in a large-scale photonic recurrent neural network,” Optica, vol. 5, no. 6, pp. 756– 760, 6 2018.
  13. [13]Hughes TW, Minkov M, Shi Y, and Fan S, “ Training of photonic neural networks through in situ backpropagation and gradient measurement,” Optica, vol. 5, no. 7, pp. 864– 871, 7 2018.
  14. [14]Shastri BJ, Tait AN, de Lima TF, Nahmias MA, Peng H-T, and Prucnal PR, “ Principles of Neuromorphic Photonics,” ArXiv180100016 Phys., pp. 1– 37. 2018.
  15. [16]Chang J, Sitzmann V, Dun X, Heidrich W, and Wetzstein G, “ Hybrid optical-electronic convolutional neural networks with optimized diffractive optics for image classification,” Sci. Rep, vol. 8, no. 1 p. 12324 8 2018. 30120316 PMC6098044
  16. [17]Soures N, Steidle J, Preble S, and Kudithipudi D, “ Neuro-MMI: A Hybrid Photonic-Electronic Machine Learning Platform,” in 2018 IEEE Photonics Society Summer Topical Meeting Series (SUM), 2018, pp. 187– 188.
  17. [18]Chakraborty I, Saha G, Sengupta A, and Roy K, “ Toward Fast Neural Computing using All-Photonic Phase Change Spiking Neurons,” Sci. Rep, vol. 8, no. 1, p. 12980, 8 2018. 30154507 PMC6113276
  18. [20]Tait AN, “ Neuromorphic photonic networks using silicon photonic weight banks,” Sci. Rep, vol. 7, no. 1, 12 2017. PMC5547135 28784997
  19. [23]LeCun Y, Bengio Y, and Hinton G, “ Deep learning,” Nature, vol. 521 no. 7553 pp. 436– 444. 5 2015. 26017442
  20. [24]Schmidhuber J, “ Deep learning in neural networks: An overview,” Neural Netw., vol. 61, no. Supplement C, pp. 85– 117, 1 2015. 25462637
  21. [25]Lecun Y, Bottou L, Bengio Y, and Haffner P, “ Gradient-based learning applied to document recognition,” Proc. IEEE, vol. 86, no. 11, pp. 2278– 2324. 11 1998.
  22. [27]Maas AL, Hannun AY, and Ng AY, “ Rectifier nonlinearities improve neural network acoustic models,” in Proc. icml, 2013, vol. 30, p. 3.
  23. [29]Wan L, Zeiler M, Zhang S, Le Cun Y, and Fergus R, “ Regularization of neural networks using dropconnect,” in International Conference on Machine Learning, 2013, pp. 1058– 1066.
  24. [30]Golik P, Doetsch P, and Ney H, “ Cross-entropy vs. squared error training: a theoretical and experimental comparison.,” in Interspeech, 2013 vol. 13 pp. 1756– 1760.
  25. [31]He K, Zhang X, Ren S, and Sun J, “ Identity Mappings in Deep Residual Networks,” in Computer Vision – ECCV 2016, 2016, pp. 630– 645.
  26. [32]Emons M, Obata K, Binhammer T, Ovsianikov A, Chichkov BN, and Morgner U, “ Two-photon polymerization technique with sub-50 nm resolution by sub-10 fs laser pulses,” Opt. Mater. Express, vol. 2, no. 7 p. 942 7 2012.
  27. [35]Zeiler MD and Fergus R, “ Visualizing and Understanding Convolutional Networks,” in Computer Vision – ECCV 2014, vol. 8689, Fleet D, Pajdla T, Schiele B, and Tuytelaars T, Eds. Cham: Springer International Publishing, 2014, pp. 818– 833.
  28. [36]Jang M, “ Wavefront shaping with disorder-engineered metasurfaces,” Nat. Photonics, vol. 12, no. 2, pp. 84– 90, 2 2018. 29527234 PMC5842956
  29. [37]Kamali SM, Arbabi E, Arbabi A, and Faraon A, “ A review of dielectric optical metasurfaces for wavefront control,” Nanophotonics, vol. 7, no. 6, pp. 1041– 1068, 6 2018.
  30. [38]Gerke TD and Piestun R, “ Aperiodic volume optics,” Nat. Photonics, vol. 4, no. 3, pp. 188– 193, 3 2010.
  31. [39]Morizur J-F, “ Programmable unitary spatial mode manipulation,” J. Opt. Soc. Am. A, vol. 27, no. 11, p. 2524, 11 2010. 21045918
  32. [43]Horowitz M, “ 1.1 Computing’s energy problem (and what we can do about it),” in 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC), San Francisco, CA, USA, 2014, pp. 10– 14.

Cited by

Sorted by citations

Authors
Mengu, D., Luo, Y., Rivenson, Y., & Ozcan, A.
Affiliations
California NanoSystems Institute (US)
ORCID
Venue
IEEE Journal of Selected Topics in Quantum Electronics
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2020-01
Volume
26
Issue
1
Pages
1-14
ISSN
  • 1077-260X (ISSN-L)
  • 1558-4542
DOI
10.1109/jstqe.2019.2921376
PMC
PMC7677864
OpenAlex
W2895491352
Cited by
255
References
4
Citations by year
2026: 30 · 2025: 40 · 2024: 47 · 2023: 45 · 2022: 38 · 2021: 31 · 2020: 17 · 2019: 6 · 2018: 1
FWCI
10.49
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Journal of Selected Topics in Quantum Electronics (publishedVersion)
  • PubMed (publishedVersion)
  • Europe PMC (PubMed Central) (acceptedVersion)
  • PubMed Central (acceptedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • Figshare (submittedVersion, cc-by)
  • arXiv (Cornell University)
Funders
  • National Science Foundation
  • Howard Hughes Medical Institute
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Reservoir Computing
  • Physical Sciences › Physics and Astronomy › Acoustics and Ultrasonics › Random lasers and scattering media
  • Physical Sciences › Engineering › Electrical and Electronic Engineering › Advanced Memory and Neural Computing
Keywords
diffractive deep neural networks, loss function design, gradient vanishing, handwritten digit recognition, low-power optical computing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review