High-capacity computing with self-rectification nonlinear optical neural processor

Abstract

Artificial intelligence (AI) and neural networks have driven groundbreaking innovations across numerous disciplines. Optical computing offers the promise of unprecedented speed and energy efficiency in the post-Moore era; however, achieving efficient, practical nonlinear activation using all-optical approaches remains a challenge. Here, we present an optical nonlinear neural processing unit (ONNPU) that implements all-optical nonlinear activation through a self-rectification mechanism. The ONNPU architecture perfectly imitates the structure of digital neural networks, enabling seamless integration with the established deep learning ecosystem. We benchmark ONNPU across nine diverse tasks spanning decision, regression and generation, including accuracies of 98.07% on MNIST and 93.54% on Fashion-MNIST. When integrated into a 201-million-parameter Vision Transformer, ONNPU achieves 82.4% top-1 accuracy on full ImageNet classification (1,000 categories); when integrated into a 117-million-parameter decoder-only Transformer, ONNPU enables short-form story generation that outperforms GPT-2. By addressing more complex and diverse deep learning tasks, ONNPU paves the way toward practical optical machine intelligence, unleashing significant potential for high-performance optical computing.

Wavelength-Multiplexed Nonlinear Computing with a Single-Layer Diffractive Optical Processor
Yongkang Cheng, Che-Yung Shen, Yuntian Wang +2
arXiv16 Sep 2026PreprintarXiv licenceFull textApplied physics & instrumentation Optics & photonicsarXiv
QuDPy-FDGF: A Python-Based Tool for Computing Ultrafast Nonlinear Optical Responses Using Frequency Domain Green's Functions
Mathieu Desmarais, Simón Paiva-Ortega, Félix Brunell +3
arXiv21 Oct 2026PreprintarXiv licenceFull textPhysical chemistry Computational physicsarXiv
Mirror-Constrained Optical Hall Rectification in Asymmetric Gold Gratings:\\ Edelstein Conversion and a Phase-Seeded Kerr Four-Wave Mixing Instability
Michael N. Leuenberger, Richard M. Osgood III
arXiv5 Oct 2026PreprintCC BY 4.0Full textCondensed matterarXiv
Programmable nonlinear function synthesis on a photonic processor with quantum signal processing
Elaheh Karooby, Masoud Hakimi Heris, Yuan Liu +1
arXiv12 Sep 2026PreprintarXiv licenceFull textQuantum physics & quantum informationarXiv
Universal Broadband Linear Optical Transformations by an Interlaced Structured Integrated Photonic Processor
Ilya V. Kondratyev, Kseniia N. Urusova, Artem S. Argenchiev +8
arXiv8 Sep 2026PreprintarXiv licenceFull textOptics & photonicsarXiv

Cite

BibTeX RIS

BibTeX

@misc{ma2026highcapacity,
  title = {{High-capacity computing with self-rectification nonlinear optical neural processor}},
  author = {Ruicheng Ma and Siyu Dong and Yuzhi Shi and Yuchen Zhu and Hong Luo and Qiang Fu and Hadi Amata and Wolfgang Heidrich and Xiong Dun and Hongfei Jiao and Hui Zhang and Qinghua Song and Zeyong Wei and Zhanshan Wang and Ali Momeni and Romain Fleury and Xinbin Cheng},
  howpublished = {arXiv},
  year = {2026},
  month = sep,
  doi = {10.48550/arxiv.2609.19981},
  eprint = {2609.19981},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.19981},
}

RIS

TY  - UNPB
TI  - High-capacity computing with self-rectification nonlinear optical neural processor
AU  - Ruicheng Ma
AU  - Siyu Dong
AU  - Yuzhi Shi
AU  - Yuchen Zhu
AU  - Hong Luo
AU  - Qiang Fu
AU  - Hadi Amata
AU  - Wolfgang Heidrich
AU  - Xiong Dun
AU  - Hongfei Jiao
AU  - Hui Zhang
AU  - Qinghua Song
AU  - Zeyong Wei
AU  - Zhanshan Wang
AU  - Ali Momeni
AU  - Romain Fleury
AU  - Xinbin Cheng
PY  - 2026
DA  - 2026/09/17
PB  - arXiv
DO  - 10.48550/arxiv.2609.19981
UR  - https://arxiv.org/abs/2609.19981
AB  - Artificial intelligence (AI) and neural networks have driven groundbreaking innovations across numerous disciplines. Optical computing offers the promise of unprecedented speed and energy efficiency in the post-Moore era; however, achieving efficient, practical nonlinear activation using all-optical approaches remains a challenge. Here, we present an optical nonlinear neural processing unit (ONNPU) that implements all-optical nonlinear activation through a self-rectification mechanism. The ONNPU architecture perfectly imitates the structure of digital neural networks, enabling seamless integration with the established deep learning ecosystem. We benchmark ONNPU across nine diverse tasks spanning decision, regression and generation, including accuracies of 98.07% on MNIST and 93.54% on Fashion-MNIST. When integrated into a 201-million-parameter Vision Transformer, ONNPU achieves 82.4% top-1 accuracy on full ImageNet classification (1,000 categories); when integrated into a 117-million-parameter decoder-only Transformer, ONNPU enables short-form story generation that outperforms GPT-2. By addressing more complex and diverse deep learning tasks, ONNPU paves the way toward practical optical machine intelligence, unleashing significant potential for high-performance optical computing.
ER  -

Abstract

Related

Cite