[17p-A25-2~11] 4.2 Photonics Devices, Photonic Integrated Circuit and Silicon Photonics

16:00 〜 16:30英語発表

[17p-A25-6] [JSAP-Optica Joint Symposia Invited Talk] Diffraction-based on-chip optical neural network with high computational density

〇Wencan Liu¹, Yuyao Huang¹, Run Sun¹, Tingzhao Fu², Hongwei Chen¹ (1.Tsinghua Univ., 2.National Univ. of Defense Technology)

キーワード：

optical computing、on-chip diffractive、high computational density

The rapid advancement of artificial intelligence has led to substantial progress in various fields with deep neural networks (DNNs). However, complex tasks often require increasing power consumption and greater resources of electronics. On-chip optical neural networks (ONNs) are increasingly recognized for their power efficiency, wide bandwidth, and capability for light-speed parallel processing. We proposed on-chip diffractive optical neural networks (DONNs) to offer the potential to map a larger number of neurons and connections onto optics. Furthermore, To improve the computational density and integration level, we proposed ultra-compact DONNs designed with the structure re-parameterization algorithmand experimentally verified their performance, which increased the computational density by more than an order of magnitude.