Supervisor’s Foreword Arti.cial intelligence is sweeping across the globe at an unprecedented pace, yet the underlying computational architecture that supports its development faces dual challenges of computing power and power consumption. Optical neural networks represent a highly forward-looking computational architecture, boasting signi.cant advantages such as ultra-high speed, low-power consumption, and large bandwidth. Currently, there has been extensive research in areas such as image processing and speech recognition, positioning it as a strong contender for the next generation of computational architectures. Among these, diffractive optical neural networks (DONNs), characterized by their high integration and high computing capacity, hold the promise of serving as a bridge for the perfect combination of optical computing and arti.cial intelligence.
This book systematically investigates the fundamental principles, design method, simulation, and experimental validation of on-chip integrated DONNs, covering the complete chain from theoretical foundations to practical veri.cation. The main inno-vations include: . A parameter training and mapping model for on-chip DONNs is proposed, which can analytically characterize the propagation laws of light in silicon-based slab waveguides and enable more accurate implementation of pre-trained key parameters on physical devices; . A silicon-based integrated DONN chip with low latency, low-power consumption, high integration, and high computing capacity is proposed, along with a compensation scheme for system errors in the chip, further ensuring the performance of the DONN chip system; . A design methodology for a multifunctional silicon-based integrated DONN chip system is proposed, which further enhances the computing ef.ciency of the DONN chip.
This research topic holds important theoretical signi.cance and practical value. Most of the content covered in this book has previously been published in the foremost journals or conferences within this research .eld. Despite the rapid development in
Supervisor’s Foreword
optical neural networks, understanding the fundamental technologies and principles of on-chip DONNs can provide readers with a comprehensive, systematic, and in-depth reference. For readers engaged in interdisciplinary research in optics, electronic information, arti.cial intelligence, and related .elds, this book may help them apply similar techniques or inspire them to develop novel solutions.
Beijing, China Hongwei Chen March 2025
Related Publications in This Thesis
[1] Fu T, Zang Y, Huang Y, et al. Photonic machine learning with on-chip diffractive optics[J]. Nature Communications, 2023, 14(1): 70
[2] Fu T, Zang Y, Huang H, et al. On-chip photonic diffractive optical neural network based on a spatial domain electromagnetic propagation model[J]. Optics Express, 2021, 29(20): 31924–31940
[3] Fu T, Huang Y, Sun R, et al. Integrated diffractive optical neural network with space-time interleaving[J]. Chinese Optics Letters, 2023, 21(9): 091301
占位居中
直播中，去观看