Transformer-based learned dimensional collapse for silicon photonic inverse design

Aras Arslan; Ahmet Onur Dasdemir; Furkan Aykut Sarikamis; Berna Kiraz; Alper Kiraz; Abdullah Magden; Emir Salih Magden

doi:10.1117/12.3080973

Transformer-based learned dimensional collapse for silicon photonic inverse design

Aras Arslan
, Ahmet Onur Dasdemir
, Furkan Aykut Sarikamis
, Berna Kiraz
, Alper Kiraz
, Abdullah Magden
, Emir Salih Magden^*

^*Corresponding author for this work

Department of Artificial Intelligence and Data Engineering

Koc University
Center for Image Analysis (OGAM)
Bursa Teknik University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

We present a deep learning framework that revolutionizes photonic device design by collapsing costly 3D electromagnetic simulations into fast, accurate 2D representations. Our dual-stage, Transformer-based architecture combines a rapid factorization-cached 2D FDFD solver with a U-Transformer network to reconstruct full 3D fields, while a dedicated phase module preserves phase integrity. Trained on 16,000 3D-FDTD simulations of random silicon photonic devices, our model achieves over 99.1% field-matching accuracy and enables inverse design optimizations that are over 100 times faster than traditional methods. Designed devices exhibit less than 0.5 dB transmission mismatch and more than 90% structural similarity to 3D-FDTD results across the 1.5-1.6 µm wavelength range. This scalable approach enables high-throughput, rapid, and practical design workflows for next-generation photonic components.

Original language	English
Title of host publication	Physics and Simulation of Optoelectronic Devices XXXIV
Editors	Marek Osinski, Yasuhiko Arakawa, Frederic Grillot, Frederic Grillot
Publisher	SPIE
ISBN (Electronic)	9781510696976
DOIs	https://doi.org/10.1117/12.3080973
Publication status	Published - 5 Mar 2026
Event	34th Physics and Simulation of Optoelectronic Devices - San Francisco, United States Duration: 19 Jan 2026 → 22 Jan 2026

Publication series

Name	Proceedings of SPIE - The International Society for Optical Engineering
Volume	13890
ISSN (Print)	0277-786X
ISSN (Electronic)	1996-756X

Conference

Conference	34th Physics and Simulation of Optoelectronic Devices
Country/Territory	United States
City	San Francisco
Period	19/01/26 → 22/01/26

Bibliographical note

Publisher Copyright:
© 2026 SPIE. All rights reserved.

Keywords

Electromagnetic simulations
Inverse-photonics design
Transformers

Access to Document

10.1117/12.3080973

Cite this

Arslan, A., Dasdemir, A. O., Sarikamis, F. A., Kiraz, B., Kiraz, A., Magden, A., & Magden, E. S. (2026). Transformer-based learned dimensional collapse for silicon photonic inverse design. In M. Osinski, Y. Arakawa, F. Grillot, & F. Grillot (Eds.), Physics and Simulation of Optoelectronic Devices XXXIV Article 1389010 (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13890). SPIE. https://doi.org/10.1117/12.3080973

Arslan, Aras ; Dasdemir, Ahmet Onur ; Sarikamis, Furkan Aykut et al. / Transformer-based learned dimensional collapse for silicon photonic inverse design. Physics and Simulation of Optoelectronic Devices XXXIV. editor / Marek Osinski ; Yasuhiko Arakawa ; Frederic Grillot ; Frederic Grillot. SPIE, 2026. (Proceedings of SPIE - The International Society for Optical Engineering).

@inproceedings{5a0dce5284f84184afe3c424fa2a8d5a,

title = "Transformer-based learned dimensional collapse for silicon photonic inverse design",

abstract = "We present a deep learning framework that revolutionizes photonic device design by collapsing costly 3D electromagnetic simulations into fast, accurate 2D representations. Our dual-stage, Transformer-based architecture combines a rapid factorization-cached 2D FDFD solver with a U-Transformer network to reconstruct full 3D fields, while a dedicated phase module preserves phase integrity. Trained on 16,000 3D-FDTD simulations of random silicon photonic devices, our model achieves over 99.1\% field-matching accuracy and enables inverse design optimizations that are over 100 times faster than traditional methods. Designed devices exhibit less than 0.5 dB transmission mismatch and more than 90\% structural similarity to 3D-FDTD results across the 1.5-1.6 µm wavelength range. This scalable approach enables high-throughput, rapid, and practical design workflows for next-generation photonic components.",

keywords = "Electromagnetic simulations, Inverse-photonics design, Transformers",

author = "Aras Arslan and Dasdemir, \{Ahmet Onur\} and Sarikamis, \{Furkan Aykut\} and Berna Kiraz and Alper Kiraz and Abdullah Magden and Magden, \{Emir Salih\}",

year = "2026",

month = mar,

day = "5",

doi = "10.1117/12.3080973",

language = "English",

series = "Proceedings of SPIE - The International Society for Optical Engineering",

publisher = "SPIE",

editor = "Marek Osinski and Yasuhiko Arakawa and Frederic Grillot and Frederic Grillot",

booktitle = "Physics and Simulation of Optoelectronic Devices XXXIV",

address = "United States",

}

Arslan, A, Dasdemir, AO, Sarikamis, FA, Kiraz, B, Kiraz, A, Magden, A & Magden, ES 2026, Transformer-based learned dimensional collapse for silicon photonic inverse design. in M Osinski, Y Arakawa, F Grillot & F Grillot (eds), Physics and Simulation of Optoelectronic Devices XXXIV., 1389010, Proceedings of SPIE - The International Society for Optical Engineering, vol. 13890, SPIE, 34th Physics and Simulation of Optoelectronic Devices, San Francisco, United States, 19/01/26. https://doi.org/10.1117/12.3080973

Transformer-based learned dimensional collapse for silicon photonic inverse design. / Arslan, Aras; Dasdemir, Ahmet Onur; Sarikamis, Furkan Aykut et al.
Physics and Simulation of Optoelectronic Devices XXXIV. ed. / Marek Osinski; Yasuhiko Arakawa; Frederic Grillot; Frederic Grillot. SPIE, 2026. 1389010 (Proceedings of SPIE - The International Society for Optical Engineering; Vol. 13890).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Transformer-based learned dimensional collapse for silicon photonic inverse design

AU - Arslan, Aras

AU - Dasdemir, Ahmet Onur

AU - Sarikamis, Furkan Aykut

AU - Kiraz, Berna

AU - Kiraz, Alper

AU - Magden, Abdullah

AU - Magden, Emir Salih

PY - 2026/3/5

Y1 - 2026/3/5

N2 - We present a deep learning framework that revolutionizes photonic device design by collapsing costly 3D electromagnetic simulations into fast, accurate 2D representations. Our dual-stage, Transformer-based architecture combines a rapid factorization-cached 2D FDFD solver with a U-Transformer network to reconstruct full 3D fields, while a dedicated phase module preserves phase integrity. Trained on 16,000 3D-FDTD simulations of random silicon photonic devices, our model achieves over 99.1% field-matching accuracy and enables inverse design optimizations that are over 100 times faster than traditional methods. Designed devices exhibit less than 0.5 dB transmission mismatch and more than 90% structural similarity to 3D-FDTD results across the 1.5-1.6 µm wavelength range. This scalable approach enables high-throughput, rapid, and practical design workflows for next-generation photonic components.

AB - We present a deep learning framework that revolutionizes photonic device design by collapsing costly 3D electromagnetic simulations into fast, accurate 2D representations. Our dual-stage, Transformer-based architecture combines a rapid factorization-cached 2D FDFD solver with a U-Transformer network to reconstruct full 3D fields, while a dedicated phase module preserves phase integrity. Trained on 16,000 3D-FDTD simulations of random silicon photonic devices, our model achieves over 99.1% field-matching accuracy and enables inverse design optimizations that are over 100 times faster than traditional methods. Designed devices exhibit less than 0.5 dB transmission mismatch and more than 90% structural similarity to 3D-FDTD results across the 1.5-1.6 µm wavelength range. This scalable approach enables high-throughput, rapid, and practical design workflows for next-generation photonic components.

KW - Electromagnetic simulations

KW - Inverse-photonics design

KW - Transformers

UR - https://www.scopus.com/pages/publications/105040048841

U2 - 10.1117/12.3080973

DO - 10.1117/12.3080973

M3 - Conference contribution

AN - SCOPUS:105040048841

T3 - Proceedings of SPIE - The International Society for Optical Engineering

BT - Physics and Simulation of Optoelectronic Devices XXXIV

A2 - Osinski, Marek

A2 - Arakawa, Yasuhiko

A2 - Grillot, Frederic

PB - SPIE

T2 - 34th Physics and Simulation of Optoelectronic Devices

Y2 - 19 January 2026 through 22 January 2026

ER -

Arslan A, Dasdemir AO, Sarikamis FA, Kiraz B, Kiraz A, Magden A et al. Transformer-based learned dimensional collapse for silicon photonic inverse design. In Osinski M, Arakawa Y, Grillot F, Grillot F, editors, Physics and Simulation of Optoelectronic Devices XXXIV. SPIE. 2026. 1389010. (Proceedings of SPIE - The International Society for Optical Engineering). doi: 10.1117/12.3080973

Transformer-based learned dimensional collapse for silicon photonic inverse design

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this