论文信息 - A Highly Adaptive Compilation Method to Break the Limitation of SNR on Calculation Precision in Photonic Neuromorphic System

A Highly Adaptive Compilation Method to Break the Limitation of SNR on Calculation Precision in Photonic Neuromorphic System

We propose a compilation method in photonic neuromorphic system to improve calculation precision and provide high adaptability. The simulation results show that our method significantly reduces the SNR requirement for a precision and outperforms physics-aware training in model inference accuracy under the same SNR.

Hongxiang Guo | Cen Wang | Jian Wu | Yuepeng Wu | Yi Guo

[1] Zuyuan He,et al. Digital-Analog Co-Design for Precision Compressed Integrated Photonic Convolution Neural Network , 2022, 2022 Optical Fiber Communications Conference and Exhibition (OFC).

[2] F. Yaman,et al. A silicon photonic–electronic neural network for fibre nonlinearity compensation , 2021, Nature Electronics.

[3] Logan G. Wright,et al. Deep physical neural networks trained with backpropagation , 2021, Nature.

[4] A. Boes,et al. 11 TOPS photonic convolutional accelerator for optical neural networks , 2021, Nature.

[5] Luca P. Carloni,et al. Silicon Photonics Codesign for Deep Learning , 2020, Proceedings of the IEEE.

[6] Xuan Li,et al. Parallel convolutional processing using an integrated photonic tensor core , 2021, Nature.

[7] Raghuraman Krishnamoorthi,et al. Quantizing deep convolutional networks for efficient inference: A whitepaper , 2018, ArXiv.

[8] Shanhui Fan,et al. Training of Photonic Neural Networks through In Situ Backpropagation , 2018, 2019 Conference on Lasers and Electro-Optics (CLEO).

[9] Wolfgang Schröder-Preikschat,et al. A Practitioner's Guide to Software-Based Soft-Error Mitigation Using AN-Codes , 2014, 2014 IEEE 15th International Symposium on High-Assurance Systems Engineering.

[10] Optical Fiber Communications Conference and Exhibition, OFC 2020, San Diego, CA, USA, March 8-12, 2020 , 2020, OFC.